[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75YT8JS997S2TE2CYMCGZ7B":3,"skill-same:01M75YT8JS997S2TE2CYMCGZ7B":79},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M75YT8JS997S2TE2CYMCGZ7B","atl",null,"Coordinate atl local mirrors across Jira and Confluence. USE WHEN orienting to atl, maintaining or recovering an existing mirror, or combining both services. DO NOT USE WHEN setup\u002Fonboarding, one service, a focused workflow, or codebase-only work is primary.","isukharev\u002Fatl","isukharev","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F4347261?v=4","01M75YQ67178RKPTQQKHF1V28G","Ivan Sukharev","01M75YQ68359TNKPGE26PY1Q33",7,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: atl\ndescription: Coordinate atl local mirrors across Jira and Confluence. USE WHEN orienting to atl, maintaining or recovering an existing mirror, or combining both services. DO NOT USE WHEN setup\u002Fonboarding, one service, a focused workflow, or codebase-only work is primary.\n---\n\u003C!-- Generated from skills-src\u002Fatl\u002FSKILL.md — edit the source and run 'make gen-plugins'. -->\n\n# Working with Confluence & Jira via `atl`\n\n`atl` is a Git-style CLI that mirrors Confluence pages and Jira issues to local disk in their\n**native formats** (Confluence Storage Format `.csf`; Jira wiki), so you can treat Atlassian\ncontent as part of your file working set — read it with Read\u002FGrep\u002FGlob, edit it, and push it back.\n\nThis skill orients you. For the actual command flows, use the focused skills:\n- **Confluence pages** (pull \u002F edit the `.md` view → `conf apply`, or the `.csf` directly \u002F validate \u002F\n  push under the version gate) → the `confluence` skill.\n- **Jira issues** (search \u002F pull \u002F create \u002F update \u002F transition \u002F comment \u002F link) → the `jira`\n  skill.\n- **First-time install & config** (`atl` binary, auth, backend URLs, mirror dir) →\n  run `\u002Fatl:setup`.\n- **Optional workflow personalization** (consent-gated sample reads, team defaults, private\n  structured profile) → explicitly invoke the `onboarding` skill. Runtime work should load only the\n  relevant profile slice with `atl profile show --section ...`; for render memory use\n  `--section render_defaults --service jira|confluence`. Never copy the profile into a repo.\n  Repeated discoveries may become consent-gated `profile suggest` artifacts; they are reviewed and\n  applied or rejected explicitly, never learned in the background. Saved `render_defaults` and\n  `preferences.mirror_root` remain memory until a separately approved runtime sync; compare them\n  with `atl config show` before relying on either.\n\nIf `atl` is not installed (`command -v atl` fails), tell the user to run `\u002Fatl:setup` first.\nIf setup or mirror health is uncertain, run offline `atl doctor` before\nidentity-bearing status output. Use `atl doctor --service jira|confluence` when\nonly one backend should contribute to health, and append `--remote` only when\nbackend metadata access is intended. Remote mode makes one single-attempt\nversion GET per ready selected service, with one additional bodyless Confluence\nreachability HEAD only after a missing version route, and never reads content\nor identities. Read `safety.configured_read_only`, `effective_read_only`, and\n`read_only_source` together; the source is exactly\n`flag|environment|configuration|none`. Treat emitted error-severity\n`problems[]` as a stop signal even though the qualified report remains on\nstdout.\nFor an explicitly requested version-pinned Data Center workflow, qualify its\nseparate fail-closed boundary with `atl compatibility status --remote`; never\ninfer support from ordinary `doctor`, a nearby patch, or an HTTP error shape.\nCompatibility providers accept no arbitrary endpoint\u002Fheader\u002Fpayload overrides\nand are not MCP routes.\n\nFor an unfamiliar Jira\u002FConfluence task that is not already governed by a\nreviewed exact-command workflow, query the exact offline route before loading\nbroad command references:\n\n```bash\natl capabilities --task jira\u002Fevidence -o text\natl capabilities --task jira\u002Fsetup -o text\natl capabilities --task jira\u002Fgraph-evidence -o text\natl capabilities --task jira\u002Finverse-reference -o text\natl capabilities --task confluence\u002Fattachment-discovery -o text\natl capabilities --task confluence\u002Fcomments -o text\natl capabilities --task confluence\u002Fspace-hierarchy -o text\n```\n\nThe closed task classes are `confluence\u002Fattachment-discovery`,\n`confluence\u002Fcomments`, `confluence\u002Fedit`, `confluence\u002Fevidence`,\n`confluence\u002Fmirror`, `confluence\u002Fspace-hierarchy`,\n`confluence\u002Ftable-analytics`,\n`jira\u002Fbatch-analysis`, `jira\u002Fboard-portfolio`, `jira\u002Fedit`, `jira\u002Fevidence`,\n`jira\u002Fgraph-evidence`, `jira\u002Finverse-reference`, `jira\u002Fmirror`, `jira\u002Fportfolio`, `jira\u002Fsetup`,\n`jira\u002Fstructure-planning`, and `knowledge\u002Fsearch`. The result is a small ordered set\nof stable capability ids with the real command path, backend access class,\nevidence\u002Fcompleteness semantics, and one focused skill reference. The text view\nomits output modes, bounded MCP scope, and CLI-only metadata; use default JSON\nwhen that routing decision needs them. Load only the named focused\nskill\u002Freference, then stop expanding the route once sufficient complete\nevidence is available. Use exact filters only;\nan unknown task\u002Fid is a loud not-found result, not a prompt for fuzzy guessing.\n`capabilities` is local\u002Foffline and works without valid config or credentials.\nWhen the effect boundary of one exact leaf matters, use `atl capabilities\n--effects --command \"\u003Ccommand>\"`; use `atl capabilities --effects` only when the\ncomplete catalog is actually needed. These static profiles are informational\nupper bounds across successful invocations. They neither authorize nor enforce\nexecution, and they do not replace read-only or guarded-write policy.\nThe additive `confluence\u002Fcomments` route keeps qualified list, exact thread,\nguarded preview, and guarded add separate. Its list\u002Fthread entries have narrower\nread-only MCP mappings; preview\u002Fadd are CLI-only and do not become plugin\nmutations.\n`confluence\u002Fattachment-discovery` is the singleton metadata-search route with a\nnarrower typed MCP mapping; `confluence\u002Fspace-hierarchy` is the singleton\nCLI-only bounded tree route. Do not substitute the broader\n`confluence\u002Fevidence` class for either.\nFor an exact Jira Structure id in the `jira\u002Fportfolio` route, use\n`jira structure get` for metadata qualification before a bounded view. Retain\nonly id, name, and read-only state for the decision; do not propagate owner,\npermission, saved-view, or forest transport payloads.\nFor Jira metadata discovery, use the ready `jira issue fields --metadata-only`\nsummary for classification, identifier-quality, and value-type aggregates\ninstead of recounting the returned field array.\nFor `confluence\u002Ftable-analytics`, prefer the content-free `conf table summary`\ndiscovery route before extracting a selected table. Its exact structural and\nstyle cardinalities remove the need to recount content-bearing raw cells. Bind\nan index selected from that summary to the same page revision with\n`--expected-version`. On extraction, use `returned_table_count` and\n`selection_reconciled` for selected-result cardinality; `table_count` remains\npage-wide. Treat `cell_count_reconciled:false` as a hard evidence failure: the\nsource placement\u002Fspan ledger did not agree with the expanded grid. A native\nspan above the supported cap fails before expansion and is not absence.\nFor durable Confluence mirror health, prefer the content-free `conf snapshot`\nroute before expanding individual pages with `conf diff`; use its reconciled\ncardinalities instead of manually joining status, validation, and render rows.\nTreat exit `8` during an active mirror mutation as a stop\u002Fretry signal; snapshot\ncoordinates through a shared advisory lock without creating or changing files.\nFor durable Jira mirror health, use the same content-free pattern with\n`jira snapshot`; its exact baseline\u002Fraw-snapshot\u002Fpending\u002Frender buckets are the\npreflight before identity-bearing `jira status` or issue-level repair.\nBoth services' status\u002Fsnapshot accept `[DIR]` or `--into`, never both, and fail\nwith exit 4 before config\u002Fnetwork unless the resolved root contains `.atl`.\nFor a single conflicted native object, route to `conf reconcile preview` or\n`jira reconcile preview`; use the separately mutation-classified `stage` leaf\nonly when an external tool needs exact private base\u002Ftheirs artifacts.\n\nWhen the installed plugin exposes `atl` MCP tools, prefer them for transient,\nbounded evidence reads: typed arguments remove shell construction and the\nserver registers no mutation or arbitrary-filesystem tool. Load\n[mcp.md](reference\u002Fmcp.md) for its exact twenty-five-tool route and CLI fallback\nboundary. Use bounded Structure metadata\u002Fview through MCP. For content-free\nhealth counts of an existing durable mirror, use the no-argument mirror snapshot\ntool only when the owner has configured `ATL_MIRROR_ROOT`. Continue using the\nCLI for raw Structure forest\u002Fvalues, mirror content\u002Fstatus\u002Fdiff, exports,\ndiff\u002Fplan, attachment content\u002Fdownloads\u002Fuploads, and every guarded write.\n\nIf `atl auth status` reports `mode: broker`, use only supported Broker routes.\nThe read subset includes Jira issue get with an explicit nonempty subset of\n`summary,description,updated`, and numeric-id Confluence page\nmetadata\u002FCSF\u002Foutline\u002Fsection reads. Treat `unsupported` as a closed capability\nboundary; do not retry through direct REST or request a backend PAT.\nFor Jira project pages, use `jira issue project-page` or `jira_project_issue_page`\nwith at most 15 issues and only summary\u002Fdescription. Every later cursor is freshly\nauthorized; `selection_complete:false` never proves stable project absence.\nDiscover this separate family with `broker discover --service jira --family\natl.broker.execution.v2` or private\n`atl:\u002F\u002Fbroker\u002Fdiscovery\u002Fjira\u002Fatl.broker.execution.v2`; omitted family retains v2.\nRead `atl broker discover --service jira|confluence`, or the corresponding\nprivate MCP `atl:\u002F\u002Fbroker\u002Fdiscovery\u002Fjira|confluence` resource, for current\noperation support and advisory access. Each read is fresh; never cache or\nreuse it as authorization. `access_request_required` does not submit a request\nor grant access, and every operation invocation reauthorizes independently.\nIn Broker mode, guarded Jira comments remain CLI-only. Preview the exact native\nbody to obtain a proposal hash and opaque ticket; apply once with that unchanged\nbody\u002Fhash\u002Fticket and the same writer session. Any ambiguous apply must be\nobserved with `jira issue comment outcome --operation-ticket \u003CSAME-TICKET>`\nusing the separately configured observer session, never replayed or routed to\ndirect Jira. Read [the Jira command reference](..\u002Fjira\u002Freference\u002Fcommands.md)\nfor the exact flags when this workflow is requested.\n\n## Mental model\n\nMirrored Atlassian content becomes local files you operate on like code: the bytes are the\nsubstrate, edits are diffed and pushed deliberately, and concurrent remote changes are caught by a\nversion gate. This is what makes `atl` \"AI-native\" — the agent works the content with its normal\nfile tools instead of round-tripping every read\u002Fwrite through an API.\n\nSee [mental-model.md](reference\u002Fmental-model.md) for when to reach for transient typed reads vs\ndurable atl CLI mirrors, and for the spec-driven \"living doc\" workflow where `atl` fits best.\n\nFor a long multi-source analysis, [delegation.md](reference\u002Fdelegation.md)\ndefines the optional one-child evidence pattern. Keep simple reads in the main\nthread and never delegate remote writes.\n\n**Working a ticket while coding?** [dev-loop.md](reference\u002Fdev-loop.md) is the end-to-end recipe:\ntake the ticket (`assign --me`, transition), keep it truthful while developing (progress comments,\ndescription updates, links), close with evidence, and update the linked Confluence page under the\nversion gate.\n\n## Two habits that matter most\n\n1. **Search first, read narrow, edit precise.** Don't bulk-dump everything and grep it. Use\n   `conf search` \u002F `jira issue search` (CQL\u002FJQL) to find the few relevant items.\n   Require Confluence top-level `complete:true` and Jira `page.complete:true`\n   before absence claims. Reuse a numeric Confluence search-result id directly\n   for outline\u002Fsection; resolve only URLs or unknown references. For a one-off read\n   use `conf page view \u003CID> -o text` or `jira issue view \u003CKEY> -o text` (configured Markdown, no\n   mirror artifacts); `pull` only what needs\n   editing or repeatable offline access,\n   read the rendered `.md` to locate, and edit there (merge back with `conf apply` \u002F `jira apply`),\n   opening the raw substrate only for what the md surface can't express.\n   Keep live reads slim too: `--fields` on Jira gets, `--columns` on Jira issue lists, and\n   command-native summary or compact projections before any local shaping — include Jira\n   `attachment` when you need the\n   presence\u002Fnames of files, but avoid a bare `issue get` because it drags the whole comment thread\n   into context.\n   [context-efficient-output.md](reference\u002Fcontext-efficient-output.md) gives the output-mode and\n   qualification rules plus the safe `jq` fallback; never add a pipe to a guarded exact-command\n   workflow.\n   Pull is non-destructive: local native or derived-view edits are preserved,\n   clean siblings continue, and a blocked selection exits 8 with content-free\n   `local_safety` evidence. Use `pull --dry-run` to inspect a refresh. Use\n   `--stash-local` only for an intentional native reset that first retains exact\n   bytes; `--overwrite-local` is the explicit discard path. Neither bypasses\n   Markdown or baseline-integrity failures.\n2. **`push` is the one deliberate checkpoint.** The Confluence safe loop is: pull fresh → edit → validate →\n   inspect offline semantics with `conf diff` → review `conf push --dry-run` against current remote state → push under the version gate. On a conflict, a human decides whether to\n   re-pull or force — never auto-force.\n   Confluence comment append uses its own guarded loop: read-only `conf comment\n   preview` → review the complete footer-root baseline-bound proposal → after\n   explicit approval, one guarded `conf comment add` with `--apply` and its\n   `--expected-proposal-hash`. Although `add` defaults to dry-run, it remains\n   mutating-classified. Inline create\u002Freply\u002Fresolve\u002Freopen use the separate\n   exact-pinned `conf comment mutation preview|apply` loop. Inline create takes\n   exact native-CSF body and bounded UTF-8 selection files plus a zero-based\n   occurrence; ATL binds the raw selection, applies the pinned-client\n   normalization\u002Fmasks fail-closed, derives raw-DOM geometry, and never edits\n   marker CSF. Never replay\n   `outcome_unknown`; the mutation surface remains CLI-only.\n\nFor every agent-created Bash block that must not mutate Jira, Confluence,\nauth\u002Fconfig, or profile state, make this export its first statement:\n\n```bash\nexport ATL_READ_ONLY=1\natl ...\natl ...\n```\n\nBefore planning Jira or Confluence writes, run `atl policy show`; when it is\nactive, plan only within its effective `grants`.\n\nEvery later `atl` call and child process in that shell inherits the guard unless\nit is explicitly overridden. A prefix such as `ATL_READ_ONLY=1 atl ...` protects\nonly that one process, so never use the one-command form for a multi-command\nread-only workflow. Do not remove or override the exported guard inside the\nworkflow. Passing global `--read-only` on every call remains a more repetitive\nalternative.\nExit 8 with `policy:\"read_only\"` is a deliberate safety refusal, not a retry;\nask the human before changing the launcher\u002Fconfig policy. Pulls, views, status,\nvalidation, and exports remain available.\n\nWhen date boundaries or timezone semantics matter, inspect them once explicitly\ninstead of inferring them from rendered timestamps:\n\n```bash\nexport ATL_READ_ONLY=1\natl environment inspect\n```\n\nThe command performs at most three sequential metadata GETs across configured\nJira\u002FConfluence, never JQL\u002FCQL\u002Fsearch\u002Fpage reads, and never runs automatically\nduring incremental pull. Preserve its evidence labels: Jira server offset and\nuser timezone may be `observed`, JQL is an `assumed` mapping from the user zone,\nCQL remains `unknown` unless a backend can prove it, and the Markdown display\nzone is `configured` or `default`.\n\nFor any JSON failure, branch on stable `kind` and numeric `code`, not words in\n`error`. Treat `remediation` and the closed schema-v1 `recovery` object as safe\nguidance to present, never authorization to\nretry a write or change policy automatically. Backend\u002FAPI prose cannot set\nthese classification fields. `rate_limited` \u002F `wait_before_retry` means bounded\nreplay-safe read retries were exhausted: do not immediately repeat the command\nor tool call, and never retry a write automatically.\nThe shared transport never follows redirects for POST, PUT, PATCH, or DELETE;\na mutating 3xx is an error to reconcile at the original target, not permission\nto replay the method or body at a server-selected path.\n`output_limit_exceeded` \u002F `narrow_or_raise_bound` means caller-selected\n`max_bytes` rejected the complete encoded result; treat it as no evidence and\nnarrow the selection or deliberately choose a larger allowed bound.\nFor Confluence table tools, `not_found` \u002F\n`summarize_then_select_table` means the selected 1-based index is outside the\ncontent-free table count. Summarize without a table selection, choose from that\ninventory, then extract once; do not report the page as missing.\nFor `confluence_table_extract`, `check_failed` \u002F\n`reread_table_summary_then_retry_expected_version` means the page changed after\nthe table index was selected. Re-read the content-free table summary, select\nthe table again from that revision, and extract it once with the new exact\n`expected_page_version`; do not retry the old positional selection against the\nnew revision.\nFor `confluence_page_section` or `confluence_page_sections`, `check_failed` or `not_found` \u002F\n`outline_then_select_section` means the occurrence selection is ambiguous or\nstale. Refresh the content-free outline, choose the exact heading occurrence,\nthen read that section once; do not report the page or heading as missing.\n`check_failed` \u002F `reread_outline_then_retry_expected_version` means the\n`expected_page_version` you supplied no longer matches the page; the message\ncarries only the expected and current integers. Re-read the outline, re-select\nthe occurrence there, and request the section once at the new version — do not\nretry the previous selection against the new revision.\nFor `jira_structure_view`, `not_found` or `check_failed` \u002F\n`view_then_select_subtree` means the Structure exists but its stored-folder\nselector did not resolve exactly. Read one selector-free bounded view with a\nnarrow field projection and `max_rows` sufficient for the full forest, choose\nthe folder `row_id`, and request that exact `folder_row` subtree once. Use the\nCLI if the full forest does not fit the MCP caps; do not report the Structure\nas missing or repeat the failed selector.\n`check_failed` \u002F `reread_structure_view_then_retry_expected_forest_version`\nmeans the `expected_forest_signature`\u002F`expected_forest_version` pair you supplied\ndoes not match the current forest; the message carries only the expected and\ncurrent integers. Re-read the view, re-select the subtree from that fresh\nresult, and request it once with the new pair.\nA returned pair with either member zero is non-bindable: omit both expected\ninputs, keep the selection explicitly ungated, and report that limitation.\n\nThe recommended convention keeps the mirror **outside the user's code\nrepository** at `~\u002F.atl\u002F\u003Cworkspace>\u002F`, so it is fully greppable yet never\ncommitted. The CLI uses that path only when the workspace exports\n`ATL_MIRROR_ROOT` or passes `--into`; otherwise built-in fallbacks are `mirror`\n(Confluence) and `mirror-jira` (Jira). Full rules are in\n[workflow.md](reference\u002Fworkflow.md).\n\nBefore remote work against an existing durable root, inspect its\ncontent-minimized service bindings with `atl mirror backend status \u003CROOT>`.\nFresh service-empty pulls and explicitly registered creates bind automatically;\nan unbound legacy root with service evidence requires the explicit local\npreview\u002Fapply workflow. After human approval to cross the read-only policy\nboundary, copy only the exact preview digest:\n\n```bash\nenv -u ATL_READ_ONLY atl mirror backend bind \u003CROOT> --service jira\nenv -u ATL_READ_ONLY atl mirror backend bind \u003CROOT> --service jira \\\n  --apply --expected-backend-sha256 '\u003Cexact backend_sha256 from preview>' \\\n  --confirm BIND\n```\n\nThe entire bind leaf, including write-free preview, is blocked by\n`ATL_READ_ONLY=1`. Bind loads no PAT, makes no backend request, stores only a\ntagged digest in private strict-v1 `.atl\u002Fbackend-bindings.json`, and never\nreplaces a different binding. A mismatch means use the original backend or a\nnew mirror. Remote mirror status\u002Fsnapshot\u002Fpush\u002Freconcile\u002Fplan phases fail before\nnetwork access when a required binding is missing or mismatched; offline mirror\nwork remains available. Persisted Confluence Jira-macro expansion requires a\nseparate Jira binding.\n\nWhen creating a Confluence page, copying a page, or creating a Jira issue that\nmust immediately join a durable mirror, opt in explicitly with both `--register`\nand `--into \u003CROOT>`. Jira creation still requires a separate preview followed by\nthe exact `--apply --expected-proposal-hash`; omission of `--apply` is preview,\nnever legacy creation. Omit only the registration pair for remote-only apply. Registration\nperforms one authoritative readback and saves mirror sync state only after the exact\nnative\u002Fbase\u002Fview artifacts are present. If the remote object was created but\nregistration fails, retain its stdout id\u002Fkey and exit 8 as evidence; never replay\nthe create\u002Fcopy. Preserve local files and recover only that identity with\n`conf pull --id \u003CID> --into \u003CROOT>` or\n`jira pull --jql 'key = \u003CKEY>' --limit 1 --into \u003CROOT>`.\n\n## New capabilities (cloud-CLI parity)\n\nRecent additions expand both surfaces — check the focused skills for full flag details:\n\n**Global flags (all commands):**\n- `--read-only` \u002F `ATL_READ_ONLY=1` — fail closed on every mutating command before credentials\u002Fbody\u002Fnetwork access.\n- `ATL_NO_UPDATE=1` — disable only the signed release check; it does not block\n  Jira\u002FConfluence reads. Homebrew launchers set it automatically and update\n  through `brew upgrade atl`. For a true air gap, combine it with offline\n  commands and an external network policy; `docs\u002Fnetwork-egress.md` is the\n  complete destination inventory.\n- `-o id` — print just the primary identifier(s) one per line (issue keys, page IDs) for safe piping into `xargs` or scripts. Not all commands support it; those that don't return an error.\n- `--verbose` \u002F `ATL_VERBOSE=1` — trace every HTTP request\u002Fresponse to stderr (token never logged).\n- Shell completion for fixed-value flags (e.g. `--output`, `--format`, `--status`) is registered.\n  Help and completion remain usable while global read-only policy is active.\n- `capabilities --task \u003Cexact-class> -o text` — compact offline task routing;\n  use default JSON only when the routing decision needs the omitted bounded MCP\n  scope, output modes, or explicit CLI-only metadata, and `-o id` only when\n  stable capability ids alone are sufficient.\n- `capabilities --effects [--command \"\u003Cexact leaf>\"]` — offline static effect\n  upper bounds for every executable leaf; informational only, never authority\n  or enforcement.\n- `mcp serve --service jira|confluence|offline` — standalone closed read-only\n  profiles; omission preserves the plugin's complete default inventory.\n\n**Confluence additions:** typed read-only `render.confluence.page_fields` shared\nby mirror and transient views; the `render.confluence.jira_macros=auto|off`\nsafety policy; guarded file-backed `conf page title set` and\nreview-bound `conf page move`; complete\u002Fguarded `conf page labels\nlist|add|remove`; safe same-origin page-reference\u002Fshort-link resolution and\nstructural outline\u002Fbounded-section reads;\ndedicated native `conf blog create` with strict CSF\u002FMarkdown validation;\nexplicit post-create `conf page create --register --into \u003CROOT>` and guarded\npreview\u002Fapply `conf page copy [--register --into \u003CROOT>]` from one authoritative\nreadback;\nopt-in ordered `conf pull --page-prefetch` plus a shared\n`--requests-per-second` transport boundary for ordinary CQL\u002Fspace and\ncomplete\u002Fincremental mirrors\nwhile every local write\u002Fcheckpoint remains serial;\n`conf page list --space [--status]`, caller-bounded qualified `conf space tree`,\n`conf page open --id`, guarded `conf page copy --id --title [--space]\n[--parent] [--register --into]\n[--apply --expected-version --expected-proposal-hash]`, metadata-only\n`conf attachment search`, `conf attachment {list,get,upload}`, guarded permanent\n`conf attachment delete --page-id --id\n[--apply --confirm DELETE --expected-version --expected-proposal-hash]`, `conf me`, `conf search --space\u002F--title\u002F--label\u002F--type`\nconvenience filters (no `--cql` needed), `.md` view renders internal links as `[[Title]]`.\n\n**Presentation time:** `render.display_time_zone` is one IANA zone for human\nJira\u002FConfluence Markdown dates (default `UTC`). It is independent of JQL\u002FCQL,\ndoes not alter exact JSON\u002Fnative timestamps, and is recorded in view state for\ndeterministic offline render\u002Fapply.\n\n**Jira additions:** typed `render.jira.field_views` (including opt-in editable\nrich-text sections with explicit pending state) and opt-in `epic_children` views;\nexplicit preview\u002Fapply `jira issue create [preview] --register --into \u003CROOT>`\nwith exact `--expected-proposal-hash` on apply, from one authoritative readback,\nwith state committed last and no create replay on local registration failure;\npreview-first permanent issue deletion bound to immutable id, freshness,\ncomplete permission-relative subtask evidence, and explicit cascade intent;\nvalue-free metadata and compact named issue-field inspection; qualified, filterable issue\nhistory with explicit completeness, deterministic cardinality\u002Fconsistency summary\nincluding separate missing\u002Fduplicate identity facts, a summary-only projection\nwithout raw changelog rows (explicit false is rejected), and last-field-change\nmetadata;\ntransient multi-key export to artifact-only stdout; deterministic epic evidence\ndigest and standalone refs with reconciled per-kind\u002Fper-source aggregates;\nbyte-compatible full schema-v2 and post-collection compact schema-v1 CLI and\ntyped MCP work-artifact graph with exact\nstructured Jira traversal, optional CLI-only Confluence id\u002Ftitle resolution, typed\nedges, mentions, budgets, frontier, metadata-reconciled fields, and qualified\nper-node sources (including experimental issue properties), plus explicit\nCLI and typed MCP fail-closed Jira Development project\u002Fcommit\u002Fbranch\u002FMR\nidentities that never trigger GitLab requests; MCP requires explicit\n`include_development:true`, omits Development-node URLs, and preserves the\nstable request and output profile when omitted or false; compact selects only\ncanonical URL-node identities and\u002For SCM coordinates while retaining bounds,\nincomplete sources, frontier, warnings, and reconciliation;\nCLI-only `jira issue reference search` starts from one exact GitLab project or\nConfluence page and scans an explicit caller-qualified Jira scope and source\nset under explicit bounds; exhaustive mode alone can prove absence, fast mode\nis always incomplete, strict mode emits its content-free qualified result\nbefore exit 8, and neither GitLab nor discovered URLs are requested;\ncheck\u002Fattachments\u002Frefs\u002Ftree. For a report\nor quarter review, route through the Jira skill's one-hop\n`reference\u002Fevidence-workflow.md` and stop once sufficient complete evidence is\navailable;\nguarded link suggestions and\nversioned plan apply; guarded file-backed custom-field preview\u002Fapply; labels;\ncomplete\u002Fguarded watcher list\u002Fadd\u002Fremove;\ncomplete worklog reads and guarded single-entry time logging without write\nretry (the `jira\u002Fedit` route exposes `jira.issue.worklog.list` and\n`jira.issue.worklog.add`);\nusers, planning reports, and compact exports.\n`jira export` manifests hash configured backend identity but retain\nselectors\u002Ffields verbatim and may still be private. Explicit key\u002Fid exports preserve de-duplicated\nselector order and omit missing identities; JQL exports retain backend order. Boards\u002Fsprints use the Jira Software Agile API. Structure\ncommands read metadata, forests, rows, values, issue snapshots, and offline\nexports, with `--root` subtree selection where supported. Breaking command\ngroups are `comment preview|add|list|delete` and\n`link add [preview]|list|delete [preview]`;\ncomment preview is GET-only, while add is dry-run by default and applies only\nwith its reviewed full-record\u002Fbody\u002Factor\u002Fissue proposal hash. Comment stdout is\ncontent-minimized; `applied` and `recovered` require exact advancing readback,\nand any attempted `outcome_unknown` must never be replayed. `issue transition preview` is\nalso GET-only; the parent transition command is dry-run by default and applies\nthe exact reviewed target\u002Fcomment\u002F`--field k=v` or `--field-json k=JSON` proposal only with its hash.\nConflict or unverifiable transition outcomes and `outcome_unknown` comments are\nnever replay-safe.\nLink preview children are GET-only; both parents require the exact reviewed\nhash for one bounded immutable-id write and use two `kind:link` policy targets.\n\n**Local manifests:** `atl manifest create --root DIR [--service\njira|confluence|generic]` omits credentials and raw backend identity, but retains\nfile counts, selectors, fields, paths, ATL version, and URL hashes. Caller\nmetadata is not redacted: never put credentials in those flags, and review the\nmanifest before publishing.\n\n**Offline indexer corpus:** use `atl corpus export --jira JIRA_ROOT\n--confluence CONF_ROOT --store STORE` for one or both initialized mirrors. The\ncommand skips configuration, credentials, self-update, and backend access. It\nreads pristine mirror baselines rather than ambient working native or Markdown\nedits, and stdout is content-free; the sealed store itself contains private\ntitles, text, paths, and references, so create it as an existing owner-only\n`0700` directory outside source repositories. Add `--initialize-store` only for\nthe first export into an empty root. ATL checks the complete projected member\nset before initializing or staging. Treat `readiness:partial` as incomplete\nevidence, and do not use `--allow-unreconciled` outside diagnostics. Preserve\nthe store when an error reports an unknown durable outcome; do not infer either\nrollback or successful publication.\n\nTo retire stale derived index entries, run `atl corpus diff --store STORE` only\nagainst the sealed current pointer. Its default output is content-free and a\n`tombstoned` count means absence from one compatible qualified successor, not\nbackend deletion. Request stable identities only with `--identity-artifact`\nunder an existing owner-only parent outside the sealed store; ATL refuses\nstore-contained and symlink-aliased destinations and never replaces the target.\nKeep that artifact private. The command never deletes a mirror, generation,\nindex, or remote object.\n\n**On-demand qualified corpus:** use `atl corpus build` only for caller-nominated\nJira projects and\u002For Confluence spaces when an indexer requires\n`readiness:ready`. Create the exact-`0700` root outside the source checkout,\nset `ATL_READ_ONLY=1`, and supply every finite selection, request,\nresponse-byte, generation, deadline, concurrency, and rate bound documented by\n`atl corpus build --help`. The command reads only selected configured backends,\nuses one shared transport budget, captures independent service windows, fixes\nminimal rendering with Jira-macro expansion off, and publishes only after all\ncapture receipts reconcile. Feed consumers only the sealed current generation;\nnever index `attempts\u002F`. Repeat exact options after a known returned read\nfailure. If recovery reports `phase=recover reason=outcome_unknown`, preserve\nthe root and use explicit `--restart`; do not infer or automatically replay the\nin-flight remote phase. For `phase=publish reason=outcome_unknown`, preserve the\nroot and repeat exact options without `--restart`; the verified generation may\nalready be current, so let ATL reconcile the visible pointer and active record.\nBuild and restart never grant backend-write or cleanup authority.\n\nFor optional cross-container reuse, add a separate exact-`0700`\n`--cache-root` plus all three positive `--cache-max-requests`,\n`--cache-max-response-bytes`, and `--cache-deadline` guards. Initialize that\nempty root once with `--initialize-cache`; keep it outside the disposable\nworkspace, source checkout, and downstream index. The first cache contract can\nreuse only an unchanged sole-Confluence capture with comments and attachments\noff, exact clean generator and configured-CA trust, matching principal\u002Fscope,\nselector\u002Foptions\u002Fschemas, and two equal complete metadata passes. Jira, mixed,\nchanged, system-trust-only, incomplete, or unqualified cases remain cold. The\ncache stores plaintext sealed content; require an encrypted volume when at-rest\nencryption is policy. A changed cache selection starts a new bounded target;\npreserve the workspace so ATL's one-way recovery marker can reconcile any\nolder workspace\u002Fcache transition instead of deleting or rewriting it by hand.\n\nInspect with `atl corpus cache status --store STORE` (or its exact local\n`corpus cache doctor` alias). Retention is never part of build: write a private exclusive plan with\n`atl corpus cache retention preview --store STORE --retain-predecessors N\n--plan-artifact FILE`, review its counts and private artifact, then run\n`atl corpus cache retention apply` only with the same file, emitted\n`--expected-plan-digest`, and `--apply`. Preserve cache and plan on any ambiguous\napply; repeat the exact plan for recovery. Status and replacement preview\nrefuse while plan-bound cleanup is pending. Never remove quarantine state by\nhand or treat status, a successful build, tombstones, or unsealed stages as\ndeletion authority.\n\nComments and attachment inventories are opt-in and require explicit per-item\npage\u002Fcount bounds. Native attachment bodies additionally require\n`--attachment-bodies`, one or more exact repeatable `--attachment-media-type`\nvalues, and per-item plus generation-wide byte bounds. Do not use wildcards or\nderive media types from filenames. Strict mode is the default; use\n`--allow-partial-evidence` only when the consumer accepts visibly partial\nqualification. Binary `asset` members stay separate from Markdown\u002FJSONL, and\nATL never follows narrative URLs, performs OCR, or expands archives.\n\n**Disposable private indexing:** use the checked-in corpus dev-container only\nwhen an indexer needs an on-demand qualified snapshot. Mount backend URLs,\nPATs, selectors, and an optional CA as owner-private runtime files; never put\ntheir values in the image, container spec, arguments, source checkout, or\nlogs. Keep the exact-`0700` runtime, empty index root, and source checkout\ndisjoint. The wrapper forces read-only\u002Fno-update mode, requires every finite\nbound, makes the statically validated build its first remote operation, and\nstarts ATL\u002Findexer children from explicit clean environments. It invokes the\nindexer only after `atl corpus handoff` verifies one sealed canonical document\ninventory. The handoff artifact must be outside the store and remains private.\nWhen an enclosing runtime has a trusted external capture verifier, use the\nseparate `atl corpus handoff-qualified --store STORE` command for an immediate\nConfluence-only cross-execution handoff. Do not pass hashes or treat its\noptional route artifact as authority: only success from the runtime's own\nstill-active bounded invocation permits immediate admission.\n`expires_at_millis` is a content-free wall-clock projection, never a renewal\ninput. Ordinary `corpus handoff` remains offline. A denied\u002Fexpired\u002Fambiguous\nresult retains data and may leave a non-authoritative exclusive route; never\ndelete or retry by guessing.\nGraphify is optional and not bundled: use an explicit absolute binary, never\nuse `--code-only`, and require reviewed egress plus an external network policy\nfor every endpoint that is not an exact loopback HTTP host and port.\nSee `docs\u002Fcorpus-devcontainer.md` for cold start, persistence, cleanup, and\nprivacy details.\n\n## Reacting to results\n\n`atl` prints JSON to stdout by default. Use `-o text` only where the command\ndocuments a human view; unsupported text requests fail with exit 2 and never\nfall back to JSON. The CLI signals outcomes\nthrough exit codes. Parse the JSON; map the exit code per [exit-codes.md](reference\u002Fexit-codes.md)\n(e.g. `5` = version conflict → preserve the candidate and run reconcile preview before any refresh or `--force`; `7` = not\nconfigured → run `\u002Fatl:setup`; `3` = the server rejected the token → re-`auth login` with a valid\nPAT).\n\n## Version skew (plugin vs binary)\n\nTreat a group name by itself as a help request; any unknown child or stray\npositional token is a usage failure (exit 2), never successful help output.\n\nThe plugin and the `atl` binary version together: each release ships both under one number, the\nbinary self-updates within ~6h of a release, and the plugin updates when its version changes. If a\ncommand **documented by these skills** fails as `unknown command`\u002F`unknown flag` (exit 2), don't\nimprovise a workaround — suspect skew: run `atl version` and compare its `version` with the installed\nplugin's version. Also retain the full `commit` and `build_state` in diagnostics; in a source\ncheckout, a different commit or `dirty` build can explain behavior that the release version alone\ncannot. `unknown` provenance is valid for an unstamped build and is not proof of tampering. An older\nbinary catches up on its next run (self-update applies on the following\ninvocation). Refresh the lagging plugin with the platform-specific flow:\n\nUse Claude Code's `\u002Fplugin update atl` command.\n\nRe-check the exact syntax with `--help` before retrying.\n\n## When something went wrong\n\nIf `atl` itself caused real friction — repeated failures on one operation, a forced fallback, a\nmisleading error, an unexpected refusal — offer the user to report it (never report on your own):\nwith consent, a **sanitized public issue** in `isukharev\u002Fatl`, and\u002For with separate consent a\n**detailed private case file** (`atl-feedback\u002F\u003Cdate>-\u003Cslug>.md`, kept out of VCS) that the user\ncan hand to their internal development team for reproduction and a fix. Triggers, both consent\ngates, the redaction checklist, and both templates: [feedback.md](reference\u002Ffeedback.md).\n","\u003C!-- Generated from skills-src\u002Fatl\u002FSKILL.md — edit the source and run 'make gen-plugins'. -->\n\n# Working with Confluence & Jira via `atl`\n\n`atl` is a Git-style CLI that mirrors Confluence pages and Jira issues to local disk in their\n**native formats** (Confluence Storage Format `.csf`; Jira wiki), so you can treat Atlassian\ncontent as part of your file working set — read it with Read\u002FGrep\u002FGlob, edit it, and push it back.\n\nThis skill orients you. For the actual command flows, use the focused skills:\n- **Confluence pages** (pull \u002F edit the `.md` view → `conf apply`, or the `.csf` directly \u002F validate \u002F\n  push under the version gate) → the `confluence` skill.\n- **Jira issues** (search \u002F pull \u002F create \u002F update \u002F transition \u002F comment \u002F link) → the `jira`\n  skill.\n- **First-time install & config** (`atl` binary, auth, backend URLs, mirror dir) →\n  run `\u002Fatl:setup`.\n- **Optional workflow personalization** (consent-gated sample reads, team defaults, private\n  structured profile) → explicitly invoke the `onboarding` skill. Runtime work should load only the\n  relevant profile slice with `atl profile show --section ...`; for render memory use\n  `--section render_defaults --service jira|confluence`. Never copy the profile into a repo.\n  Repeated discoveries may become consent-gated `profile suggest` artifacts; they are reviewed and\n  applied or rejected explicitly, never learned in the background. Saved `render_defaults` and\n  `preferences.mirror_root` remain memory until a separately approved runtime sync; compare them\n  with `atl config show` before relying on either.\n\nIf `atl` is not installed (`command -v atl` fails), tell the user to run `\u002Fatl:setup` first.\nIf setup or mirror health is uncertain, run offline `atl doctor` before\nidentity-bearing status output. Use `atl doctor --service jira|confluence` when\nonly one backend should contribute to health, and append `--remote` only when\nbackend metadata access is intended. Remote mode makes one single-attempt\nversion GET per ready selected service, with one additional bodyless Confluence\nreachability HEAD only after a missing version route, and never reads content\nor identities. Read `safety.configured_read_only`, `effective_read_only`, and\n`read_only_source` together; the source is exactly\n`flag|environment|configuration|none`. Treat emitted error-severity\n`problems[]` as a stop signal even though the qualified report remains on\nstdout.\nFor an explicitly requested version-pinned Data Center workflow, qualify its\nseparate fail-closed boundary with `atl compatibility status --remote`; never\ninfer support from ordinary `doctor`, a nearby patch, or an HTTP error shape.\nCompatibility providers accept no arbitrary endpoint\u002Fheader\u002Fpayload overrides\nand are not MCP routes.\n\nFor an unfamiliar Jira\u002FConfluence task that is not already governed by a\nreviewed exact-command workflow, query the exact offline route before loading\nbroad command references:\n\n```bash\natl capabilities --task jira\u002Fevidence -o text\natl capabilities --task jira\u002Fsetup -o text\natl capabilities --task jira\u002Fgraph-evidence -o text\natl capabilities --task jira\u002Finverse-reference -o text\natl capabilities --task confluence\u002Fattachment-discovery -o text\natl capabilities --task confluence\u002Fcomments -o text\natl capabilities --task confluence\u002Fspace-hierarchy -o text\n```\n\nThe closed task classes are `confluence\u002Fattachment-discovery`,\n`confluence\u002Fcomments`, `confluence\u002Fedit`, `confluence\u002Fevidence`,\n`confluence\u002Fmirror`, `confluence\u002Fspace-hierarchy`,\n`confluence\u002Ftable-analytics`,\n`jira\u002Fbatch-analysis`, `jira\u002Fboard-portfolio`, `jira\u002Fedit`, `jira\u002Fevidence`,\n`jira\u002Fgraph-evidence`, `jira\u002Finverse-reference`, `jira\u002Fmirror`, `jira\u002Fportfolio`, `jira\u002Fsetup`,\n`jira\u002Fstructure-planning`, and `knowledge\u002Fsearch`. The result is a small ordered set\nof stable capability ids with the real command path, backend access class,\nevidence\u002Fcompleteness semantics, and one focused skill reference. The text view\nomits output modes, bounded MCP scope, and CLI-only metadata; use default JSON\nwhen that routing decision needs them. Load only the named focused\nskill\u002Freference, then stop expanding the route once sufficient complete\nevidence is available. Use exact filters only;\nan unknown task\u002Fid is a loud not-found result, not a prompt for fuzzy guessing.\n`capabilities` is local\u002Foffline and works without valid config or credentials.\nWhen the effect boundary of one exact leaf matters, use `atl capabilities\n--effects --command \"\u003Ccommand>\"`; use `atl capabilities --effects` only when the\ncomplete catalog is actually needed. These static profiles are informational\nupper bounds across successful invocations. They neither authorize nor enforce\nexecution, and they do not replace read-only or guarded-write policy.\nThe additive `confluence\u002Fcomments` route keeps qualified list, exact thread,\nguarded preview, and guarded add separate. Its list\u002Fthread entries have narrower\nread-only MCP mappings; preview\u002Fadd are CLI-only and do not become plugin\nmutations.\n`confluence\u002Fattachment-discovery` is the singleton metadata-search route with a\nnarrower typed MCP mapping; `confluence\u002Fspace-hierarchy` is the singleton\nCLI-only bounded tree route. Do not substitute the broader\n`confluence\u002Fevidence` class for either.\nFor an exact Jira Structure id in the `jira\u002Fportfolio` route, use\n`jira structure get` for metadata qualification before a bounded view. Retain\nonly id, name, and read-only state for the decision; do not propagate owner,\npermission, saved-view, or forest transport payloads.\nFor Jira metadata discovery, use the ready `jira issue fields --metadata-only`\nsummary for classification, identifier-quality, and value-type aggregates\ninstead of recounting the returned field array.\nFor `confluence\u002Ftable-analytics`, prefer the content-free `conf table summary`\ndiscovery route before extracting a selected table. Its exact structural and\nstyle cardinalities remove the need to recount content-bearing raw cells. Bind\nan index selected from that summary to the same page revision with\n`--expected-version`. On extraction, use `returned_table_count` and\n`selection_reconciled` for selected-result cardinality; `table_count` remains\npage-wide. Treat `cell_count_reconciled:false` as a hard evidence failure: the\nsource placement\u002Fspan ledger did not agree with the expanded grid. A native\nspan above the supported cap fails before expansion and is not absence.\nFor durable Confluence mirror health, prefer the content-free `conf snapshot`\nroute before expanding individual pages with `conf diff`; use its reconciled\ncardinalities instead of manually joining status, validation, and render rows.\nTreat exit `8` during an active mirror mutation as a stop\u002Fretry signal; snapshot\ncoordinates through a shared advisory lock without creating or changing files.\nFor durable Jira mirror health, use the same content-free pattern with\n`jira snapshot`; its exact baseline\u002Fraw-snapshot\u002Fpending\u002Frender buckets are the\npreflight before identity-bearing `jira status` or issue-level repair.\nBoth services' status\u002Fsnapshot accept `[DIR]` or `--into`, never both, and fail\nwith exit 4 before config\u002Fnetwork unless the resolved root contains `.atl`.\nFor a single conflicted native object, route to `conf reconcile preview` or\n`jira reconcile preview`; use the separately mutation-classified `stage` leaf\nonly when an external tool needs exact private base\u002Ftheirs artifacts.\n\nWhen the installed plugin exposes `atl` MCP tools, prefer them for transient,\nbounded evidence reads: typed arguments remove shell construction and the\nserver registers no mutation or arbitrary-filesystem tool. Load\n[mcp.md](reference\u002Fmcp.md) for its exact twenty-five-tool route and CLI fallback\nboundary. Use bounded Structure metadata\u002Fview through MCP. For content-free\nhealth counts of an existing durable mirror, use the no-argument mirror snapshot\ntool only when the owner has configured `ATL_MIRROR_ROOT`. Continue using the\nCLI for raw Structure forest\u002Fvalues, mirror content\u002Fstatus\u002Fdiff, exports,\ndiff\u002Fplan, attachment content\u002Fdownloads\u002Fuploads, and every guarded write.\n\nIf `atl auth status` reports `mode: broker`, use only supported Broker routes.\nThe read subset includes Jira issue get with an explicit nonempty subset of\n`summary,description,updated`, and numeric-id Confluence page\nmetadata\u002FCSF\u002Foutline\u002Fsection reads. Treat `unsupported` as a closed capability\nboundary; do not retry through direct REST or request a backend PAT.\nFor Jira project pages, use `jira issue project-page` or `jira_project_issue_page`\nwith at most 15 issues and only summary\u002Fdescription. Every later cursor is freshly\nauthorized; `selection_complete:false` never proves stable project absence.\nDiscover this separate family with `broker discover --service jira --family\natl.broker.execution.v2` or private\n`atl:\u002F\u002Fbroker\u002Fdiscovery\u002Fjira\u002Fatl.broker.execution.v2`; omitted family retains v2.\nRead `atl broker discover --service jira|confluence`, or the corresponding\nprivate MCP `atl:\u002F\u002Fbroker\u002Fdiscovery\u002Fjira|confluence` resource, for current\noperation support and advisory access. Each read is fresh; never cache or\nreuse it as authorization. `access_request_required` does not submit a request\nor grant access, and every operation invocation reauthorizes independently.\nIn Broker mode, guarded Jira comments remain CLI-only. Preview the exact native\nbody to obtain a proposal hash and opaque ticket; apply once with that unchanged\nbody\u002Fhash\u002Fticket and the same writer session. Any ambiguous apply must be\nobserved with `jira issue comment outcome --operation-ticket \u003CSAME-TICKET>`\nusing the separately configured observer session, never replayed or routed to\ndirect Jira. Read [the Jira command reference](..\u002Fjira\u002Freference\u002Fcommands.md)\nfor the exact flags when this workflow is requested.\n\n## Mental model\n\nMirrored Atlassian content becomes local files you operate on like code: the bytes are the\nsubstrate, edits are diffed and pushed deliberately, and concurrent remote changes are caught by a\nversion gate. This is what makes `atl` \"AI-native\" — the agent works the content with its normal\nfile tools instead of round-tripping every read\u002Fwrite through an API.\n\nSee [mental-model.md](reference\u002Fmental-model.md) for when to reach for transient typed reads vs\ndurable atl CLI mirrors, and for the spec-driven \"living doc\" workflow where `atl` fits best.\n\nFor a long multi-source analysis, [delegation.md](reference\u002Fdelegation.md)\ndefines the optional one-child evidence pattern. Keep simple reads in the main\nthread and never delegate remote writes.\n\n**Working a ticket while coding?** [dev-loop.md](reference\u002Fdev-loop.md) is the end-to-end recipe:\ntake the ticket (`assign --me`, transition), keep it truthful while developing (progress comments,\ndescription updates, links), close with evidence, and update the linked Confluence page under the\nversion gate.\n\n## Two habits that matter most\n\n1. **Search first, read narrow, edit precise.** Don't bulk-dump everything and grep it. Use\n   `conf search` \u002F `jira issue search` (CQL\u002FJQL) to find the few relevant items.\n   Require Confluence top-level `complete:true` and Jira `page.complete:true`\n   before absence claims. Reuse a numeric Confluence search-result id directly\n   for outline\u002Fsection; resolve only URLs or unknown references. For a one-off read\n   use `conf page view \u003CID> -o text` or `jira issue view \u003CKEY> -o text` (configured Markdown, no\n   mirror artifacts); `pull` only what needs\n   editing or repeatable offline access,\n   read the rendered `.md` to locate, and edit there (merge back with `conf apply` \u002F `jira apply`),\n   opening the raw substrate only for what the md surface can't express.\n   Keep live reads slim too: `--fields` on Jira gets, `--columns` on Jira issue lists, and\n   command-native summary or compact projections before any local shaping — include Jira\n   `attachment` when you need the\n   presence\u002Fnames of files, but avoid a bare `issue get` because it drags the whole comment thread\n   into context.\n   [context-efficient-output.md](reference\u002Fcontext-efficient-output.md) gives the output-mode and\n   qualification rules plus the safe `jq` fallback; never add a pipe to a guarded exact-command\n   workflow.\n   Pull is non-destructive: local native or derived-view edits are preserved,\n   clean siblings continue, and a blocked selection exits 8 with content-free\n   `local_safety` evidence. Use `pull --dry-run` to inspect a refresh. Use\n   `--stash-local` only for an intentional native reset that first retains exact\n   bytes; `--overwrite-local` is the explicit discard path. Neither bypasses\n   Markdown or baseline-integrity failures.\n2. **`push` is the one deliberate checkpoint.** The Confluence safe loop is: pull fresh → edit → validate →\n   inspect offline semantics with `conf diff` → review `conf push --dry-run` against current remote state → push under the version gate. On a conflict, a human decides whether to\n   re-pull or force — never auto-force.\n   Confluence comment append uses its own guarded loop: read-only `conf comment\n   preview` → review the complete footer-root baseline-bound proposal → after\n   explicit approval, one guarded `conf comment add` with `--apply` and its\n   `--expected-proposal-hash`. Although `add` defaults to dry-run, it remains\n   mutating-classified. Inline create\u002Freply\u002Fresolve\u002Freopen use the separate\n   exact-pinned `conf comment mutation preview|apply` loop. Inline create takes\n   exact native-CSF body and bounded UTF-8 selection files plus a zero-based\n   occurrence; ATL binds the raw selection, applies the pinned-client\n   normalization\u002Fmasks fail-closed, derives raw-DOM geometry, and never edits\n   marker CSF. Never replay\n   `outcome_unknown`; the mutation surface remains CLI-only.\n\nFor every agent-created Bash block that must not mutate Jira, Confluence,\nauth\u002Fconfig, or profile state, make this export its first statement:\n\n```bash\nexport ATL_READ_ONLY=1\natl ...\natl ...\n```\n\nBefore planning Jira or Confluence writes, run `atl policy show`; when it is\nactive, plan only within its effective `grants`.\n\nEvery later `atl` call and child process in that shell inherits the guard unless\nit is explicitly overridden. A prefix such as `ATL_READ_ONLY=1 atl ...` protects\nonly that one process, so never use the one-command form for a multi-command\nread-only workflow. Do not remove or override the exported guard inside the\nworkflow. Passing global `--read-only` on every call remains a more repetitive\nalternative.\nExit 8 with `policy:\"read_only\"` is a deliberate safety refusal, not a retry;\nask the human before changing the launcher\u002Fconfig policy. Pulls, views, status,\nvalidation, and exports remain available.\n\nWhen date boundaries or timezone semantics matter, inspect them once explicitly\ninstead of inferring them from rendered timestamps:\n\n```bash\nexport ATL_READ_ONLY=1\natl environment inspect\n```\n\nThe command performs at most three sequential metadata GETs across configured\nJira\u002FConfluence, never JQL\u002FCQL\u002Fsearch\u002Fpage reads, and never runs automatically\nduring incremental pull. Preserve its evidence labels: Jira server offset and\nuser timezone may be `observed`, JQL is an `assumed` mapping from the user zone,\nCQL remains `unknown` unless a backend can prove it, and the Markdown display\nzone is `configured` or `default`.\n\nFor any JSON failure, branch on stable `kind` and numeric `code`, not words in\n`error`. Treat `remediation` and the closed schema-v1 `recovery` object as safe\nguidance to present, never authorization to\nretry a write or change policy automatically. Backend\u002FAPI prose cannot set\nthese classification fields. `rate_limited` \u002F `wait_before_retry` means bounded\nreplay-safe read retries were exhausted: do not immediately repeat the command\nor tool call, and never retry a write automatically.\nThe shared transport never follows redirects for POST, PUT, PATCH, or DELETE;\na mutating 3xx is an error to reconcile at the original target, not permission\nto replay the method or body at a server-selected path.\n`output_limit_exceeded` \u002F `narrow_or_raise_bound` means caller-selected\n`max_bytes` rejected the complete encoded result; treat it as no evidence and\nnarrow the selection or deliberately choose a larger allowed bound.\nFor Confluence table tools, `not_found` \u002F\n`summarize_then_select_table` means the selected 1-based index is outside the\ncontent-free table count. Summarize without a table selection, choose from that\ninventory, then extract once; do not report the page as missing.\nFor `confluence_table_extract`, `check_failed` \u002F\n`reread_table_summary_then_retry_expected_version` means the page changed after\nthe table index was selected. Re-read the content-free table summary, select\nthe table again from that revision, and extract it once with the new exact\n`expected_page_version`; do not retry the old positional selection against the\nnew revision.\nFor `confluence_page_section` or `confluence_page_sections`, `check_failed` or `not_found` \u002F\n`outline_then_select_section` means the occurrence selection is ambiguous or\nstale. Refresh the content-free outline, choose the exact heading occurrence,\nthen read that section once; do not report the page or heading as missing.\n`check_failed` \u002F `reread_outline_then_retry_expected_version` means the\n`expected_page_version` you supplied no longer matches the page; the message\ncarries only the expected and current integers. Re-read the outline, re-select\nthe occurrence there, and request the section once at the new version — do not\nretry the previous selection against the new revision.\nFor `jira_structure_view`, `not_found` or `check_failed` \u002F\n`view_then_select_subtree` means the Structure exists but its stored-folder\nselector did not resolve exactly. Read one selector-free bounded view with a\nnarrow field projection and `max_rows` sufficient for the full forest, choose\nthe folder `row_id`, and request that exact `folder_row` subtree once. Use the\nCLI if the full forest does not fit the MCP caps; do not report the Structure\nas missing or repeat the failed selector.\n`check_failed` \u002F `reread_structure_view_then_retry_expected_forest_version`\nmeans the `expected_forest_signature`\u002F`expected_forest_version` pair you supplied\ndoes not match the current forest; the message carries only the expected and\ncurrent integers. Re-read the view, re-select the subtree from that fresh\nresult, and request it once with the new pair.\nA returned pair with either member zero is non-bindable: omit both expected\ninputs, keep the selection explicitly ungated, and report that limitation.\n\nThe recommended convention keeps the mirror **outside the user's code\nrepository** at `~\u002F.atl\u002F\u003Cworkspace>\u002F`, so it is fully greppable yet never\ncommitted. The CLI uses that path only when the workspace exports\n`ATL_MIRROR_ROOT` or passes `--into`; otherwise built-in fallbacks are `mirror`\n(Confluence) and `mirror-jira` (Jira). Full rules are in\n[workflow.md](reference\u002Fworkflow.md).\n\nBefore remote work against an existing durable root, inspect its\ncontent-minimized service bindings with `atl mirror backend status \u003CROOT>`.\nFresh service-empty pulls and explicitly registered creates bind automatically;\nan unbound legacy root with service evidence requires the explicit local\npreview\u002Fapply workflow. After human approval to cross the read-only policy\nboundary, copy only the exact preview digest:\n\n```bash\nenv -u ATL_READ_ONLY atl mirror backend bind \u003CROOT> --service jira\nenv -u ATL_READ_ONLY atl mirror backend bind \u003CROOT> --service jira \\\n  --apply --expected-backend-sha256 '\u003Cexact backend_sha256 from preview>' \\\n  --confirm BIND\n```\n\nThe entire bind leaf, including write-free preview, is blocked by\n`ATL_READ_ONLY=1`. Bind loads no PAT, makes no backend request, stores only a\ntagged digest in private strict-v1 `.atl\u002Fbackend-bindings.json`, and never\nreplaces a different binding. A mismatch means use the original backend or a\nnew mirror. Remote mirror status\u002Fsnapshot\u002Fpush\u002Freconcile\u002Fplan phases fail before\nnetwork access when a required binding is missing or mismatched; offline mirror\nwork remains available. Persisted Confluence Jira-macro expansion requires a\nseparate Jira binding.\n\nWhen creating a Confluence page, copying a page, or creating a Jira issue that\nmust immediately join a durable mirror, opt in explicitly with both `--register`\nand `--into \u003CROOT>`. Jira creation still requires a separate preview followed by\nthe exact `--apply --expected-proposal-hash`; omission of `--apply` is preview,\nnever legacy creation. Omit only the registration pair for remote-only apply. Registration\nperforms one authoritative readback and saves mirror sync state only after the exact\nnative\u002Fbase\u002Fview artifacts are present. If the remote object was created but\nregistration fails, retain its stdout id\u002Fkey and exit 8 as evidence; never replay\nthe create\u002Fcopy. Preserve local files and recover only that identity with\n`conf pull --id \u003CID> --into \u003CROOT>` or\n`jira pull --jql 'key = \u003CKEY>' --limit 1 --into \u003CROOT>`.\n\n## New capabilities (cloud-CLI parity)\n\nRecent additions expand both surfaces — check the focused skills for full flag details:\n\n**Global flags (all commands):**\n- `--read-only` \u002F `ATL_READ_ONLY=1` — fail closed on every mutating command before credentials\u002Fbody\u002Fnetwork access.\n- `ATL_NO_UPDATE=1` — disable only the signed release check; it does not block\n  Jira\u002FConfluence reads. Homebrew launchers set it automatically and update\n  through `brew upgrade atl`. For a true air gap, combine it with offline\n  commands and an external network policy; `docs\u002Fnetwork-egress.md` is the\n  complete destination inventory.\n- `-o id` — print just the primary identifier(s) one per line (issue keys, page IDs) for safe piping into `xargs` or scripts. Not all commands support it; those that don't return an error.\n- `--verbose` \u002F `ATL_VERBOSE=1` — trace every HTTP request\u002Fresponse to stderr (token never logged).\n- Shell completion for fixed-value flags (e.g. `--output`, `--format`, `--status`) is registered.\n  Help and completion remain usable while global read-only policy is active.\n- `capabilities --task \u003Cexact-class> -o text` — compact offline task routing;\n  use default JSON only when the routing decision needs the omitted bounded MCP\n  scope, output modes, or explicit CLI-only metadata, and `-o id` only when\n  stable capability ids alone are sufficient.\n- `capabilities --effects [--command \"\u003Cexact leaf>\"]` — offline static effect\n  upper bounds for every executable leaf; informational only, never authority\n  or enforcement.\n- `mcp serve --service jira|confluence|offline` — standalone closed read-only\n  profiles; omission preserves the plugin's complete default inventory.\n\n**Confluence additions:** typed read-only `render.confluence.page_fields` shared\nby mirror and transient views; the `render.confluence.jira_macros=auto|off`\nsafety policy; guarded file-backed `conf page title set` and\nreview-bound `conf page move`; complete\u002Fguarded `conf page labels\nlist|add|remove`; safe same-origin page-reference\u002Fshort-link resolution and\nstructural outline\u002Fbounded-section reads;\ndedicated native `conf blog create` with strict CSF\u002FMarkdown validation;\nexplicit post-create `conf page create --register --into \u003CROOT>` and guarded\npreview\u002Fapply `conf page copy [--register --into \u003CROOT>]` from one authoritative\nreadback;\nopt-in ordered `conf pull --page-prefetch` plus a shared\n`--requests-per-second` transport boundary for ordinary CQL\u002Fspace and\ncomplete\u002Fincremental mirrors\nwhile every local write\u002Fcheckpoint remains serial;\n`conf page list --space [--status]`, caller-bounded qualified `conf space tree`,\n`conf page open --id`, guarded `conf page copy --id --title [--space]\n[--parent] [--register --into]\n[--apply --expected-version --expected-proposal-hash]`, metadata-only\n`conf attachment search`, `conf attachment {list,get,upload}`, guarded permanent\n`conf attachment delete --page-id --id\n[--apply --confirm DELETE --expected-version --expected-proposal-hash]`, `conf me`, `conf search --space\u002F--title\u002F--label\u002F--type`\nconvenience filters (no `--cql` needed), `.md` view renders internal links as `[[Title]]`.\n\n**Presentation time:** `render.display_time_zone` is one IANA zone for human\nJira\u002FConfluence Markdown dates (default `UTC`). It is independent of JQL\u002FCQL,\ndoes not alter exact JSON\u002Fnative timestamps, and is recorded in view state for\ndeterministic offline render\u002Fapply.\n\n**Jira additions:** typed `render.jira.field_views` (including opt-in editable\nrich-text sections with explicit pending state) and opt-in `epic_children` views;\nexplicit preview\u002Fapply `jira issue create [preview] --register --into \u003CROOT>`\nwith exact `--expected-proposal-hash` on apply, from one authoritative readback,\nwith state committed last and no create replay on local registration failure;\npreview-first permanent issue deletion bound to immutable id, freshness,\ncomplete permission-relative subtask evidence, and explicit cascade intent;\nvalue-free metadata and compact named issue-field inspection; qualified, filterable issue\nhistory with explicit completeness, deterministic cardinality\u002Fconsistency summary\nincluding separate missing\u002Fduplicate identity facts, a summary-only projection\nwithout raw changelog rows (explicit false is rejected), and last-field-change\nmetadata;\ntransient multi-key export to artifact-only stdout; deterministic epic evidence\ndigest and standalone refs with reconciled per-kind\u002Fper-source aggregates;\nbyte-compatible full schema-v2 and post-collection compact schema-v1 CLI and\ntyped MCP work-artifact graph with exact\nstructured Jira traversal, optional CLI-only Confluence id\u002Ftitle resolution, typed\nedges, mentions, budgets, frontier, metadata-reconciled fields, and qualified\nper-node sources (including experimental issue properties), plus explicit\nCLI and typed MCP fail-closed Jira Development project\u002Fcommit\u002Fbranch\u002FMR\nidentities that never trigger GitLab requests; MCP requires explicit\n`include_development:true`, omits Development-node URLs, and preserves the\nstable request and output profile when omitted or false; compact selects only\ncanonical URL-node identities and\u002For SCM coordinates while retaining bounds,\nincomplete sources, frontier, warnings, and reconciliation;\nCLI-only `jira issue reference search` starts from one exact GitLab project or\nConfluence page and scans an explicit caller-qualified Jira scope and source\nset under explicit bounds; exhaustive mode alone can prove absence, fast mode\nis always incomplete, strict mode emits its content-free qualified result\nbefore exit 8, and neither GitLab nor discovered URLs are requested;\ncheck\u002Fattachments\u002Frefs\u002Ftree. For a report\nor quarter review, route through the Jira skill's one-hop\n`reference\u002Fevidence-workflow.md` and stop once sufficient complete evidence is\navailable;\nguarded link suggestions and\nversioned plan apply; guarded file-backed custom-field preview\u002Fapply; labels;\ncomplete\u002Fguarded watcher list\u002Fadd\u002Fremove;\ncomplete worklog reads and guarded single-entry time logging without write\nretry (the `jira\u002Fedit` route exposes `jira.issue.worklog.list` and\n`jira.issue.worklog.add`);\nusers, planning reports, and compact exports.\n`jira export` manifests hash configured backend identity but retain\nselectors\u002Ffields verbatim and may still be private. Explicit key\u002Fid exports preserve de-duplicated\nselector order and omit missing identities; JQL exports retain backend order. Boards\u002Fsprints use the Jira Software Agile API. Structure\ncommands read metadata, forests, rows, values, issue snapshots, and offline\nexports, with `--root` subtree selection where supported. Breaking command\ngroups are `comment preview|add|list|delete` and\n`link add [preview]|list|delete [preview]`;\ncomment preview is GET-only, while add is dry-run by default and applies only\nwith its reviewed full-record\u002Fbody\u002Factor\u002Fissue proposal hash. Comment stdout is\ncontent-minimized; `applied` and `recovered` require exact advancing readback,\nand any attempted `outcome_unknown` must never be replayed. `issue transition preview` is\nalso GET-only; the parent transition command is dry-run by default and applies\nthe exact reviewed target\u002Fcomment\u002F`--field k=v` or `--field-json k=JSON` proposal only with its hash.\nConflict or unverifiable transition outcomes and `outcome_unknown` comments are\nnever replay-safe.\nLink preview children are GET-only; both parents require the exact reviewed\nhash for one bounded immutable-id write and use two `kind:link` policy targets.\n\n**Local manifests:** `atl manifest create --root DIR [--service\njira|confluence|generic]` omits credentials and raw backend identity, but retains\nfile counts, selectors, fields, paths, ATL version, and URL hashes. Caller\nmetadata is not redacted: never put credentials in those flags, and review the\nmanifest before publishing.\n\n**Offline indexer corpus:** use `atl corpus export --jira JIRA_ROOT\n--confluence CONF_ROOT --store STORE` for one or both initialized mirrors. The\ncommand skips configuration, credentials, self-update, and backend access. It\nreads pristine mirror baselines rather than ambient working native or Markdown\nedits, and stdout is content-free; the sealed store itself contains private\ntitles, text, paths, and references, so create it as an existing owner-only\n`0700` directory outside source repositories. Add `--initialize-store` only for\nthe first export into an empty root. ATL checks the complete projected member\nset before initializing or staging. Treat `readiness:partial` as incomplete\nevidence, and do not use `--allow-unreconciled` outside diagnostics. Preserve\nthe store when an error reports an unknown durable outcome; do not infer either\nrollback or successful publication.\n\nTo retire stale derived index entries, run `atl corpus diff --store STORE` only\nagainst the sealed current pointer. Its default output is content-free and a\n`tombstoned` count means absence from one compatible qualified successor, not\nbackend deletion. Request stable identities only with `--identity-artifact`\nunder an existing owner-only parent outside the sealed store; ATL refuses\nstore-contained and symlink-aliased destinations and never replaces the target.\nKeep that artifact private. The command never deletes a mirror, generation,\nindex, or remote object.\n\n**On-demand qualified corpus:** use `atl corpus build` only for caller-nominated\nJira projects and\u002For Confluence spaces when an indexer requires\n`readiness:ready`. Create the exact-`0700` root outside the source checkout,\nset `ATL_READ_ONLY=1`, and supply every finite selection, request,\nresponse-byte, generation, deadline, concurrency, and rate bound documented by\n`atl corpus build --help`. The command reads only selected configured backends,\nuses one shared transport budget, captures independent service windows, fixes\nminimal rendering with Jira-macro expansion off, and publishes only after all\ncapture receipts reconcile. Feed consumers only the sealed current generation;\nnever index `attempts\u002F`. Repeat exact options after a known returned read\nfailure. If recovery reports `phase=recover reason=outcome_unknown`, preserve\nthe root and use explicit `--restart`; do not infer or automatically replay the\nin-flight remote phase. For `phase=publish reason=outcome_unknown`, preserve the\nroot and repeat exact options without `--restart`; the verified generation may\nalready be current, so let ATL reconcile the visible pointer and active record.\nBuild and restart never grant backend-write or cleanup authority.\n\nFor optional cross-container reuse, add a separate exact-`0700`\n`--cache-root` plus all three positive `--cache-max-requests`,\n`--cache-max-response-bytes`, and `--cache-deadline` guards. Initialize that\nempty root once with `--initialize-cache`; keep it outside the disposable\nworkspace, source checkout, and downstream index. The first cache contract can\nreuse only an unchanged sole-Confluence capture with comments and attachments\noff, exact clean generator and configured-CA trust, matching principal\u002Fscope,\nselector\u002Foptions\u002Fschemas, and two equal complete metadata passes. Jira, mixed,\nchanged, system-trust-only, incomplete, or unqualified cases remain cold. The\ncache stores plaintext sealed content; require an encrypted volume when at-rest\nencryption is policy. A changed cache selection starts a new bounded target;\npreserve the workspace so ATL's one-way recovery marker can reconcile any\nolder workspace\u002Fcache transition instead of deleting or rewriting it by hand.\n\nInspect with `atl corpus cache status --store STORE` (or its exact local\n`corpus cache doctor` alias). Retention is never part of build: write a private exclusive plan with\n`atl corpus cache retention preview --store STORE --retain-predecessors N\n--plan-artifact FILE`, review its counts and private artifact, then run\n`atl corpus cache retention apply` only with the same file, emitted\n`--expected-plan-digest`, and `--apply`. Preserve cache and plan on any ambiguous\napply; repeat the exact plan for recovery. Status and replacement preview\nrefuse while plan-bound cleanup is pending. Never remove quarantine state by\nhand or treat status, a successful build, tombstones, or unsealed stages as\ndeletion authority.\n\nComments and attachment inventories are opt-in and require explicit per-item\npage\u002Fcount bounds. Native attachment bodies additionally require\n`--attachment-bodies`, one or more exact repeatable `--attachment-media-type`\nvalues, and per-item plus generation-wide byte bounds. Do not use wildcards or\nderive media types from filenames. Strict mode is the default; use\n`--allow-partial-evidence` only when the consumer accepts visibly partial\nqualification. Binary `asset` members stay separate from Markdown\u002FJSONL, and\nATL never follows narrative URLs, performs OCR, or expands archives.\n\n**Disposable private indexing:** use the checked-in corpus dev-container only\nwhen an indexer needs an on-demand qualified snapshot. Mount backend URLs,\nPATs, selectors, and an optional CA as owner-private runtime files; never put\ntheir values in the image, container spec, arguments, source checkout, or\nlogs. Keep the exact-`0700` runtime, empty index root, and source checkout\ndisjoint. The wrapper forces read-only\u002Fno-update mode, requires every finite\nbound, makes the statically validated build its first remote operation, and\nstarts ATL\u002Findexer children from explicit clean environments. It invokes the\nindexer only after `atl corpus handoff` verifies one sealed canonical document\ninventory. The handoff artifact must be outside the store and remains private.\nWhen an enclosing runtime has a trusted external capture verifier, use the\nseparate `atl corpus handoff-qualified --store STORE` command for an immediate\nConfluence-only cross-execution handoff. Do not pass hashes or treat its\noptional route artifact as authority: only success from the runtime's own\nstill-active bounded invocation permits immediate admission.\n`expires_at_millis` is a content-free wall-clock projection, never a renewal\ninput. Ordinary `corpus handoff` remains offline. A denied\u002Fexpired\u002Fambiguous\nresult retains data and may leave a non-authoritative exclusive route; never\ndelete or retry by guessing.\nGraphify is optional and not bundled: use an explicit absolute binary, never\nuse `--code-only`, and require reviewed egress plus an external network policy\nfor every endpoint that is not an exact loopback HTTP host and port.\nSee `docs\u002Fcorpus-devcontainer.md` for cold start, persistence, cleanup, and\nprivacy details.\n\n## Reacting to results\n\n`atl` prints JSON to stdout by default. Use `-o text` only where the command\ndocuments a human view; unsupported text requests fail with exit 2 and never\nfall back to JSON. The CLI signals outcomes\nthrough exit codes. Parse the JSON; map the exit code per [exit-codes.md](reference\u002Fexit-codes.md)\n(e.g. `5` = version conflict → preserve the candidate and run reconcile preview before any refresh or `--force`; `7` = not\nconfigured → run `\u002Fatl:setup`; `3` = the server rejected the token → re-`auth login` with a valid\nPAT).\n\n## Version skew (plugin vs binary)\n\nTreat a group name by itself as a help request; any unknown child or stray\npositional token is a usage failure (exit 2), never successful help output.\n\nThe plugin and the `atl` binary version together: each release ships both under one number, the\nbinary self-updates within ~6h of a release, and the plugin updates when its version changes. If a\ncommand **documented by these skills** fails as `unknown command`\u002F`unknown flag` (exit 2), don't\nimprovise a workaround — suspect skew: run `atl version` and compare its `version` with the installed\nplugin's version. Also retain the full `commit` and `build_state` in diagnostics; in a source\ncheckout, a different commit or `dirty` build can explain behavior that the release version alone\ncannot. `unknown` provenance is valid for an unstamped build and is not proof of tampering. An older\nbinary catches up on its next run (self-update applies on the following\ninvocation). Refresh the lagging plugin with the platform-specific flow:\n\nUse Claude Code's `\u002Fplugin update atl` command.\n\nRe-check the exact syntax with `--help` before retrying.\n\n## When something went wrong\n\nIf `atl` itself caused real friction — repeated failures on one operation, a forced fallback, a\nmisleading error, an unexpected refusal — offer the user to report it (never report on your own):\nwith consent, a **sanitized public issue** in `isukharev\u002Fatl`, and\u002For with separate consent a\n**detailed private case file** (`atl-feedback\u002F\u003Cdate>-\u003Cslug>.md`, kept out of VCS) that the user\ncan hand to their internal development team for reproduction and a fix. Triggers, both consent\ngates, the redaction checklist, and both templates: [feedback.md](reference\u002Ffeedback.md).\n",{"name":5,"description":7},"main","c6ed32ed384aa5e9a3e86d4528f16dc13833ce66","skills\u002Fatl",1,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl","https:\u002F\u002Fgithub.com\u002Fisukharev","2026-09-14 23:56:52",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add isukharev\u002Fatl --skill atl",{"tool":37,"command":38},"bunx","bunx skills add isukharev\u002Fatl --skill atl",{"tool":40,"command":41},"pnpm","pnpm dlx skills add isukharev\u002Fatl --skill atl","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002FSKILL.md",true,[45,51,55,59,63,67,71,75],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"reference\u002Fcontext-efficient-output.md","REFERENCE","参考文档",7289,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fcontext-efficient-output.md",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"reference\u002Fdelegation.md",2059,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fdelegation.md",{"path":56,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":57,"sourceHtmlUrl":58},"reference\u002Fdev-loop.md",6261,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fdev-loop.md",{"path":60,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":61,"sourceHtmlUrl":62},"reference\u002Fexit-codes.md",3282,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fexit-codes.md",{"path":64,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":65,"sourceHtmlUrl":66},"reference\u002Ffeedback.md",4712,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Ffeedback.md",{"path":68,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":69,"sourceHtmlUrl":70},"reference\u002Fmcp.md",25190,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fmcp.md",{"path":72,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":73,"sourceHtmlUrl":74},"reference\u002Fmental-model.md",2763,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fmental-model.md",{"path":76,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":77,"sourceHtmlUrl":78},"reference\u002Fworkflow.md",6959,"https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fatl\u002Freference\u002Fworkflow.md",[80,87,94,101,108,115],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":19},"01M75YVB9DFRMQ0P4SB1RWCH8D","triage-issue","Triage a bug against Jira with atl before proposing a ticket. USE WHEN an error or stack trace comes with duplicate, prior-fix, or file-a-bug intent. DO NOT USE WHEN debugging code only, inspecting a known issue, or creating an ordinary task.","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Ftriage-issue\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":19},"01M75YV7K5C8A8NV1SYHM413B7","status-report","Draft an audience-oriented Jira status report with atl. USE WHEN the user wants a weekly\u002Fdaily progress narrative, blocker summary, or period report. DO NOT USE WHEN the output is a visual dashboard, direct Jira query, knowledge search, or codebase summary.","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fstatus-report\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":19},"01M75YV57J2G080TFCV6QYWK2B","sprint-dashboard","Build a visual read-only Jira sprint or board dashboard with atl. USE WHEN the user wants a standup view, WIP picture, or sprint snapshot. DO NOT USE WHEN the output is a narrative report, raw Jira list, or codebase dashboard.","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fsprint-dashboard\u002FSKILL.md",[],[],{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":19},"01M75YV2JNBA9PTQA06JJSZGW5","spec-to-backlog","Convert a specification into a reviewed Jira epic and backlog with atl. USE WHEN decomposing requirements, RFCs, or designs into linked Jira work. DO NOT USE WHEN the source is meeting notes, only a summary is wanted, or the task is a direct service operation.","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fspec-to-backlog\u002FSKILL.md",[],[],{"skillNo":109,"name":110,"nameZh":6,"description":111,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":112,"categories":113,"keywords":114,"publishedAt":19},"01M75YTWKKWHKGZ0XQACM7VXNR","setup","Install and configure atl authentication, backends, and mirror defaults. USE WHEN the user explicitly requests install, authentication, setup repair, or \u002Fatl:setup. DO NOT USE WHEN handling normal Jira, Confluence, search, reporting, or mirror work; explicit-only.","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fsetup\u002FSKILL.md",[],[],{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":19},"01M75YTSX670SW6NSBA1HGAG65","search-knowledge","Answer from qualified Jira and Confluence evidence with atl. USE WHEN organizational knowledge is unknown or split across both services. DO NOT USE WHEN one exact resource is known, the task is a report\u002Fdashboard, or the answer belongs in the codebase.","https:\u002F\u002Fgithub.com\u002Fisukharev\u002Fatl\u002Fblob\u002Fmain\u002Fskills\u002Fsearch-knowledge\u002FSKILL.md",[],[]]