[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7DBRXT98J0Q84K38FCFQ9VT":3,"skill-same:01M7DBRXT98J0Q84K38FCFQ9VT":46},{"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":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M7DBRXT98J0Q84K38FCFQ9VT","work-authoring",null,"Draft a pi-work workspec when an outcome, scope, or proof obligation is still unclear. Use this before turning a conversation, issue, or design into a markdown draft, especially when acceptance criteria need to survive lossless promotion into strict YAML.","CenterForAgenticAI\u002Ftools","CenterForAgenticAI","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F250424195?v=4","01M7DBNBBRNZ6NT6WYSEPYK811","CAAIR","01M7DBNBCBH3B7HHWP0ACQW3F7",0,"MIT","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-work\u002Fskills\u002Fwork-authoring\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: work-authoring\ndescription: >-\n  Draft a pi-work workspec when an outcome, scope, or proof obligation is still\n  unclear. Use this before turning a conversation, issue, or design into a\n  markdown draft, especially when acceptance criteria need to survive lossless\n  promotion into strict YAML.\ndisable-model-invocation: false\n---\n\n# Work authoring\n\nUse this skill when the work is still being shaped. The output of this skill is\nan honest draft and a small set of observable acceptance criteria. It is not a\nstatus record, a dispatch plan, or a promise that the work is complete.\n\nLoad and reuse `grilling` by name when the request is vague or has meaningful\ntradeoffs. Load and reuse `domain-modeling` by name when the outcome has several\nactors, states, inputs, or boundaries.\nThose skills own the interview and modeling methods; this skill owns the\npi-work draft envelope and criteria discipline. Do not copy those skills into\nthis one.\n\n## Keep the layers separate\n\nA pi-work draft is Markdown. A promoted workspec is pure YAML. Promotion carries\nmarked content into the strict schema; it does not prove the work, run a worker,\nor create lifecycle state.\n\nKeep these layers distinct:\n\n- **Draft** — a conversational document with up to four marked regions: summary, rationale, criteria, and optional decisions.\n- **Workspec** — structured intent, a work graph, node-owned criteria, and proof\n  obligations. It has no `status`, `done`, lifecycle, or review-approval field.\n- **Execution** — the session agent decides ordering and uses pi-delegate for\n  workers. pi-work owns no scheduler or loop.\n- **Verification** — `work_verify` runs authored evidence against an explicit\n  tree and commit. A cache or dispatch receipt is not verification.\n\nDo not add a spec-level requirements list or a `satisfies` mapping. Acceptance\ncriteria belong to exactly one node. A cross-node requirement needs an explicit\nintegration node with its own criterion.\n\n## 1. Interview the outcome\n\nBefore writing prose, establish the smallest useful model. Ask one question at\na time when the answer changes the work graph or proof:\n\n1. What outcome should exist, for whom, and how will they recognize it?\n2. What is in scope, what is explicitly out of scope, and what must not change?\n3. What inputs, constraints, dependencies, and external actions matter?\n4. Which result is observable in the target tree, and which result needs a user\n   confirmation?\n5. What would make the outcome unsafe to call complete?\n6. Which decisions are unresolved, who decides them, and at what boundary must\n   they be settled?\n\nUse domain modeling to name actors and boundaries rather than inventing\nimplementation nodes from nouns in the request. Separate a deliverable from\nan activity, and separate an activity from its evidence. If an answer cannot be\nknown yet, record it as an open decision in the eventual workspec; do not hide\nit in a confident sentence.\n\nA good intent says why the work exists, its constraints, and its non-goals. A\ngood description says what outcome the workspec represents. They are related but\nnot the same paragraph.\n\n## 2. Write the marked draft\n\nUse `.work\u002Fdrafts\u002F\u003Cname>.md` unless the repository has an established draft\nlocation. The promotion markers must be the final content of column-zero ATX\nheadings. Drafts may contain up to four marked regions: summary, rationale,\ncriteria, and optional decisions. The headings for regions that are present\nmust be siblings at one level. Summary, rationale and criteria are the common\nset; add decisions when the work has questions that are not yet answered:\n\n```markdown\n# Ship the accessibility report\n\n## Summary \u003C!-- work:summary -->\nPublish an accessible report with a reproducible source record and a human\nconfirmation that the published copy is readable.\n\n## Why \u003C!-- work:rationale -->\nThe report is required for the release review. The source record makes the\nnumbers traceable; the non-goals are a redesign of the reporting pipeline and\nan automated claim about a human reader's judgment.\n\n## Open decisions \u003C!-- work:decisions -->\n- publication-target: Which channel publishes the report?\n  Two are plausible and they need different accessibility checks.\n  tripwire: Before the publishing node is dispatched.\n  decides: user\n\n## Acceptance criteria \u003C!-- work:criteria -->\n- source-record: The source record contains the measured inputs and the report build emits `source-records`.\n- published-copy: A human confirms that the published copy is the intended report and is readable.\n```\n\nThe exact criteria envelope is:\n\n- one or more top-level bullets;\n- `- \u003Cid>: \u003Cstatement>` with an id matching\n  `[A-Za-z0-9][A-Za-z0-9._-]*`;\n- unique ids within this criteria region;\n- a non-empty statement; and\n- optional continuation lines indented by at least two spaces or one tab.\n\nThe decisions envelope is the same bullet shape with two required fields:\n\n- `- \u003Cid>: \u003Cquestion>` with the same id pattern, unique within the region;\n- a non-empty question;\n- optional indented continuation lines, which extend the question; and\n- exactly one indented `tripwire: \u003Cwhen it must be answered>` line and exactly\n  one indented `decides: \u003Cwho answers it>` line, both after any continuation.\n\n`work_promote` turns each entry into an `open_decisions` record. A decision\nwithout a tripwire is a question nobody is told when to answer, so promotion\nrefuses the whole draft rather than emitting one. Recording the question is\nthe point: an unanswered question belongs in this region, never dissolved into\na confident sentence in the rationale.\n\nA promoted decision has no `gates`, so it blocks every node until it is\nresolved. A draft has no node graph to name yet; scope a decision to the nodes\nit blocks by adding `gates` during decomposition.\n\nTwo cautions the parser will not save you from. An indented line that is not\n`tripwire:` or `decides:` is treated as question prose, so an invented field\nsuch as `owner: finance` is silently appended to the question text — write\nonly the two fields this envelope defines. And a draft that puts its decisions\nunder a heading without the `\u003C!-- work:decisions -->` marker promotes cleanly\nwith no decisions at all, because an unmarked heading is ordinary prose.\n\nDo not put the marker in a paragraph, code fence, nested heading, or heading\nwith trailing prose after the marker. The region ends at the next heading of\nthe same or higher level. Keep the criteria region byte-stable: promotion is\nlossless and the criterion statements are the source text that later reviewers\nsee.\n\nUse EARS-style phrasing as writing discipline, not as a schema field. EARS\nmeans expressing a requirement in terms of a condition, an actor or system,\nan observable action, and an expected result. Prefer:\n\n```text\nWhen the report is built from the release inputs, the build emits\n`source-records` and writes the named artifact.\n```\n\nover:\n\n```text\nThe report works correctly.\n```\n\nDo not add `pattern`, `ears`, `satisfies`, or a requirement enum. The schema\nstores the active `id`, `statement`, one evidence object, and an optional\nrecorded amendment chain. Criteria quality is judgment owned by this skill and\nthe later reviewer, not a guessed parser.\n\n### One criterion, one check\n\nA criterion carries exactly one evidence object, so a criterion that fuses two\ndifferent checks cannot express evidence for either. Split it.\n\nThe tell is that the halves would be proved differently. `npm test` runs the\ncompiled suite **and** a past mutant run was observed failing is two criteria:\nthe first is a `command`, the second is a `user` confirmation of something\nalready witnessed. Written as one, both collapse to `user`, the weakest kind,\nand the machine-checkable half stops being checked.\n\nSplit on the evidence, not on the sentence. Two clauses proved by the same\ncommand are one criterion; one clause needing two different proofs is two.\n\nThis is not enforced mechanically. Detecting it means guessing from wording,\nand most criteria legitimately contain \"and\", so a check would flag honest\ncriteria and train its reader to ignore it. It is caught here and at review.\n\nSplitting a criterion that is already promoted changes the draft too: every\ncriterion must match its draft text or carry a recorded amendment chain, so\namend the original through `work_amend_criterion` and add the new half to the\ndraft's criteria region.\n\n## 3. Choose evidence honestly\n\nEvery promoted criterion needs one evidence kind:\n\n- `command` has `run` and `expect.exit`. Add a non-empty\n  `expect.output_includes` positive output floor whenever the command must prove\n  an observed effect. The current schema still accepts an absent or empty floor,\n  but `work_verify` fails closed for it; never write a criterion that only passes\n  because a command exited 0. Optionally set `timeout_ms` alongside `run`\n  and `expect` when a gate needs more than the 30,000 ms default; it must be\n  an integer from 1,000 to 3,600,000 ms (one hour). Invalid values fail\n  validation rather than being clamped. Optionally add `inherit_env: [VARIABLE_NAME]`\n  to explicitly pass named host variables to the command; unset names fail closed\n  and exact inherited values are redacted from rendered output, details,\n  failures, and cache. Values shorter than 8 characters, and output floors\n  that contain or are contained in an inherited value, fail closed by name.\n  Redaction does not contain commands that transform or transmit secrets; only\n  inherit secrets into commands you trust. Do not put values in the workspec.\n  Draft Markdown criteria are preserved verbatim by `work_promote`; add\n  `timeout_ms` to the evidence object when\n  decomposing the promoted skeleton.\n- `agent` names an agent, its named `inputs`, and a `rubric`. In v1 this returns\n  the typed `agent-unavailable` outcome and fails closed because pi-work's\n  trusted adapter for the required pi-delegate programmatic-client contract is\n  not available. It is unavailable, not skipped. Do not inject a judge, call a\n  private verifier, or treat an agent's prose as trusted evidence. Use `user`\n  evidence for judgment work in v1.\n- `user` has a `prompt`. It requires an explicit human confirmation through the\n  host UI or an exact challenge in the real session file. Silence, an assistant\n  statement, or a cache record is not confirmation.\n\nUse `!md` for prose block scalars and `!bash` for shell evidence in the promoted\nYAML. These tags are presentation only; they do not alter semantics. Other\nallowlisted scalar tags are also syntax highlighting, not evidence authority.\n\n## 4. Promote without claiming completion\n\nBefore promotion, check that the draft contains the required criteria marker.\n`work_promote` carries that criteria block verbatim, retains the visible source-draft\npath, and fails loudly when the block is missing or malformed. `work_validate`\nre-reads that draft and derives each criterion's SHA-256 root from its statement;\nno root value is authored in the workspec. Optional summary and rationale regions seed\n`description` and `intent`; missing optional regions are reported explicitly,\nnot silently invented. If the host command can prompt for missing optional\nregions, the command may do so; the non-interactive tool must not pretend that a\nuser supplied them.\n\nAfter promotion, inspect the YAML rather than assuming the skeleton is sound:\n\n- `title`, `description`, `intent`, and `work` are present;\n- the first generated comment visibly names the source draft used for lineage;\n- every node has an `id` and `task`;\n- `work` and `checklist` are mutually exclusive on each node;\n- `depends_on` names siblings only;\n- each criterion is on one owning node and has exactly one evidence object;\n- there is no lifecycle\u002Fstatus field; and\n- each command criterion has a meaningful expected effect.\n\nRun `work_validate({ path: \"\u003Cspec-path>\" })` and read its typed result. A\nwarning is not an error, but a valid result is not proof that the work happened.\nKeep the source YAML and its validation output available for the decomposition\nstep.\n\n## Authoring limits in v1\n\n- Do not promise `verified-this-session` from a draft, a promoted file, a\n  dispatch receipt, or a cache hit.\n- `work_status --refresh` is blocked in v1: refresh cannot run evidence for you.\n  `work_status` reports `verified-this-session` only after `work_verify` verifies\n  that node in this process against the exact source, worktree, and commit.\n  Everything else reports an observation state or `unverified`.\n- A source-level `work_validate` or `work_promote` export can be exercised by\n  tests or a direct host integration, but do not claim a tool ran unless the\n  host actually registered and invoked it.\n- Review approval is reporting data, not part of v1 `done`. A node is done only\n  from current-session evidence green plus full checklist accounting.\n- A green evidence result is narrow: it proves the authored evidence ran through\n  the authoritative verifier for the explicit tree and commit and that the\n  verifier accepted its concrete proof and checklist accounting. It does not\n  prove hermetic execution, reproducibility, adequate criteria, independent\n  truth of checklist or external-action claims, review approval, or durable\n  authority after serialization.\n\nIf the request needs a capability that is not available, record the limitation\nor an `open_decisions` entry. Never convert an aspiration into a criterion that\nthe v1 tools cannot execute or prove.\n","# Work authoring\n\nUse this skill when the work is still being shaped. The output of this skill is\nan honest draft and a small set of observable acceptance criteria. It is not a\nstatus record, a dispatch plan, or a promise that the work is complete.\n\nLoad and reuse `grilling` by name when the request is vague or has meaningful\ntradeoffs. Load and reuse `domain-modeling` by name when the outcome has several\nactors, states, inputs, or boundaries.\nThose skills own the interview and modeling methods; this skill owns the\npi-work draft envelope and criteria discipline. Do not copy those skills into\nthis one.\n\n## Keep the layers separate\n\nA pi-work draft is Markdown. A promoted workspec is pure YAML. Promotion carries\nmarked content into the strict schema; it does not prove the work, run a worker,\nor create lifecycle state.\n\nKeep these layers distinct:\n\n- **Draft** — a conversational document with up to four marked regions: summary, rationale, criteria, and optional decisions.\n- **Workspec** — structured intent, a work graph, node-owned criteria, and proof\n  obligations. It has no `status`, `done`, lifecycle, or review-approval field.\n- **Execution** — the session agent decides ordering and uses pi-delegate for\n  workers. pi-work owns no scheduler or loop.\n- **Verification** — `work_verify` runs authored evidence against an explicit\n  tree and commit. A cache or dispatch receipt is not verification.\n\nDo not add a spec-level requirements list or a `satisfies` mapping. Acceptance\ncriteria belong to exactly one node. A cross-node requirement needs an explicit\nintegration node with its own criterion.\n\n## 1. Interview the outcome\n\nBefore writing prose, establish the smallest useful model. Ask one question at\na time when the answer changes the work graph or proof:\n\n1. What outcome should exist, for whom, and how will they recognize it?\n2. What is in scope, what is explicitly out of scope, and what must not change?\n3. What inputs, constraints, dependencies, and external actions matter?\n4. Which result is observable in the target tree, and which result needs a user\n   confirmation?\n5. What would make the outcome unsafe to call complete?\n6. Which decisions are unresolved, who decides them, and at what boundary must\n   they be settled?\n\nUse domain modeling to name actors and boundaries rather than inventing\nimplementation nodes from nouns in the request. Separate a deliverable from\nan activity, and separate an activity from its evidence. If an answer cannot be\nknown yet, record it as an open decision in the eventual workspec; do not hide\nit in a confident sentence.\n\nA good intent says why the work exists, its constraints, and its non-goals. A\ngood description says what outcome the workspec represents. They are related but\nnot the same paragraph.\n\n## 2. Write the marked draft\n\nUse `.work\u002Fdrafts\u002F\u003Cname>.md` unless the repository has an established draft\nlocation. The promotion markers must be the final content of column-zero ATX\nheadings. Drafts may contain up to four marked regions: summary, rationale,\ncriteria, and optional decisions. The headings for regions that are present\nmust be siblings at one level. Summary, rationale and criteria are the common\nset; add decisions when the work has questions that are not yet answered:\n\n```markdown\n# Ship the accessibility report\n\n## Summary \u003C!-- work:summary -->\nPublish an accessible report with a reproducible source record and a human\nconfirmation that the published copy is readable.\n\n## Why \u003C!-- work:rationale -->\nThe report is required for the release review. The source record makes the\nnumbers traceable; the non-goals are a redesign of the reporting pipeline and\nan automated claim about a human reader's judgment.\n\n## Open decisions \u003C!-- work:decisions -->\n- publication-target: Which channel publishes the report?\n  Two are plausible and they need different accessibility checks.\n  tripwire: Before the publishing node is dispatched.\n  decides: user\n\n## Acceptance criteria \u003C!-- work:criteria -->\n- source-record: The source record contains the measured inputs and the report build emits `source-records`.\n- published-copy: A human confirms that the published copy is the intended report and is readable.\n```\n\nThe exact criteria envelope is:\n\n- one or more top-level bullets;\n- `- \u003Cid>: \u003Cstatement>` with an id matching\n  `[A-Za-z0-9][A-Za-z0-9._-]*`;\n- unique ids within this criteria region;\n- a non-empty statement; and\n- optional continuation lines indented by at least two spaces or one tab.\n\nThe decisions envelope is the same bullet shape with two required fields:\n\n- `- \u003Cid>: \u003Cquestion>` with the same id pattern, unique within the region;\n- a non-empty question;\n- optional indented continuation lines, which extend the question; and\n- exactly one indented `tripwire: \u003Cwhen it must be answered>` line and exactly\n  one indented `decides: \u003Cwho answers it>` line, both after any continuation.\n\n`work_promote` turns each entry into an `open_decisions` record. A decision\nwithout a tripwire is a question nobody is told when to answer, so promotion\nrefuses the whole draft rather than emitting one. Recording the question is\nthe point: an unanswered question belongs in this region, never dissolved into\na confident sentence in the rationale.\n\nA promoted decision has no `gates`, so it blocks every node until it is\nresolved. A draft has no node graph to name yet; scope a decision to the nodes\nit blocks by adding `gates` during decomposition.\n\nTwo cautions the parser will not save you from. An indented line that is not\n`tripwire:` or `decides:` is treated as question prose, so an invented field\nsuch as `owner: finance` is silently appended to the question text — write\nonly the two fields this envelope defines. And a draft that puts its decisions\nunder a heading without the `\u003C!-- work:decisions -->` marker promotes cleanly\nwith no decisions at all, because an unmarked heading is ordinary prose.\n\nDo not put the marker in a paragraph, code fence, nested heading, or heading\nwith trailing prose after the marker. The region ends at the next heading of\nthe same or higher level. Keep the criteria region byte-stable: promotion is\nlossless and the criterion statements are the source text that later reviewers\nsee.\n\nUse EARS-style phrasing as writing discipline, not as a schema field. EARS\nmeans expressing a requirement in terms of a condition, an actor or system,\nan observable action, and an expected result. Prefer:\n\n```text\nWhen the report is built from the release inputs, the build emits\n`source-records` and writes the named artifact.\n```\n\nover:\n\n```text\nThe report works correctly.\n```\n\nDo not add `pattern`, `ears`, `satisfies`, or a requirement enum. The schema\nstores the active `id`, `statement`, one evidence object, and an optional\nrecorded amendment chain. Criteria quality is judgment owned by this skill and\nthe later reviewer, not a guessed parser.\n\n### One criterion, one check\n\nA criterion carries exactly one evidence object, so a criterion that fuses two\ndifferent checks cannot express evidence for either. Split it.\n\nThe tell is that the halves would be proved differently. `npm test` runs the\ncompiled suite **and** a past mutant run was observed failing is two criteria:\nthe first is a `command`, the second is a `user` confirmation of something\nalready witnessed. Written as one, both collapse to `user`, the weakest kind,\nand the machine-checkable half stops being checked.\n\nSplit on the evidence, not on the sentence. Two clauses proved by the same\ncommand are one criterion; one clause needing two different proofs is two.\n\nThis is not enforced mechanically. Detecting it means guessing from wording,\nand most criteria legitimately contain \"and\", so a check would flag honest\ncriteria and train its reader to ignore it. It is caught here and at review.\n\nSplitting a criterion that is already promoted changes the draft too: every\ncriterion must match its draft text or carry a recorded amendment chain, so\namend the original through `work_amend_criterion` and add the new half to the\ndraft's criteria region.\n\n## 3. Choose evidence honestly\n\nEvery promoted criterion needs one evidence kind:\n\n- `command` has `run` and `expect.exit`. Add a non-empty\n  `expect.output_includes` positive output floor whenever the command must prove\n  an observed effect. The current schema still accepts an absent or empty floor,\n  but `work_verify` fails closed for it; never write a criterion that only passes\n  because a command exited 0. Optionally set `timeout_ms` alongside `run`\n  and `expect` when a gate needs more than the 30,000 ms default; it must be\n  an integer from 1,000 to 3,600,000 ms (one hour). Invalid values fail\n  validation rather than being clamped. Optionally add `inherit_env: [VARIABLE_NAME]`\n  to explicitly pass named host variables to the command; unset names fail closed\n  and exact inherited values are redacted from rendered output, details,\n  failures, and cache. Values shorter than 8 characters, and output floors\n  that contain or are contained in an inherited value, fail closed by name.\n  Redaction does not contain commands that transform or transmit secrets; only\n  inherit secrets into commands you trust. Do not put values in the workspec.\n  Draft Markdown criteria are preserved verbatim by `work_promote`; add\n  `timeout_ms` to the evidence object when\n  decomposing the promoted skeleton.\n- `agent` names an agent, its named `inputs`, and a `rubric`. In v1 this returns\n  the typed `agent-unavailable` outcome and fails closed because pi-work's\n  trusted adapter for the required pi-delegate programmatic-client contract is\n  not available. It is unavailable, not skipped. Do not inject a judge, call a\n  private verifier, or treat an agent's prose as trusted evidence. Use `user`\n  evidence for judgment work in v1.\n- `user` has a `prompt`. It requires an explicit human confirmation through the\n  host UI or an exact challenge in the real session file. Silence, an assistant\n  statement, or a cache record is not confirmation.\n\nUse `!md` for prose block scalars and `!bash` for shell evidence in the promoted\nYAML. These tags are presentation only; they do not alter semantics. Other\nallowlisted scalar tags are also syntax highlighting, not evidence authority.\n\n## 4. Promote without claiming completion\n\nBefore promotion, check that the draft contains the required criteria marker.\n`work_promote` carries that criteria block verbatim, retains the visible source-draft\npath, and fails loudly when the block is missing or malformed. `work_validate`\nre-reads that draft and derives each criterion's SHA-256 root from its statement;\nno root value is authored in the workspec. Optional summary and rationale regions seed\n`description` and `intent`; missing optional regions are reported explicitly,\nnot silently invented. If the host command can prompt for missing optional\nregions, the command may do so; the non-interactive tool must not pretend that a\nuser supplied them.\n\nAfter promotion, inspect the YAML rather than assuming the skeleton is sound:\n\n- `title`, `description`, `intent`, and `work` are present;\n- the first generated comment visibly names the source draft used for lineage;\n- every node has an `id` and `task`;\n- `work` and `checklist` are mutually exclusive on each node;\n- `depends_on` names siblings only;\n- each criterion is on one owning node and has exactly one evidence object;\n- there is no lifecycle\u002Fstatus field; and\n- each command criterion has a meaningful expected effect.\n\nRun `work_validate({ path: \"\u003Cspec-path>\" })` and read its typed result. A\nwarning is not an error, but a valid result is not proof that the work happened.\nKeep the source YAML and its validation output available for the decomposition\nstep.\n\n## Authoring limits in v1\n\n- Do not promise `verified-this-session` from a draft, a promoted file, a\n  dispatch receipt, or a cache hit.\n- `work_status --refresh` is blocked in v1: refresh cannot run evidence for you.\n  `work_status` reports `verified-this-session` only after `work_verify` verifies\n  that node in this process against the exact source, worktree, and commit.\n  Everything else reports an observation state or `unverified`.\n- A source-level `work_validate` or `work_promote` export can be exercised by\n  tests or a direct host integration, but do not claim a tool ran unless the\n  host actually registered and invoked it.\n- Review approval is reporting data, not part of v1 `done`. A node is done only\n  from current-session evidence green plus full checklist accounting.\n- A green evidence result is narrow: it proves the authored evidence ran through\n  the authoritative verifier for the explicit tree and commit and that the\n  verifier accepted its concrete proof and checklist accounting. It does not\n  prove hermetic execution, reproducibility, adequate criteria, independent\n  truth of checklist or external-action claims, review approval, or durable\n  authority after serialization.\n\nIf the request needs a capability that is not available, record the limitation\nor an `open_decisions` entry. Never convert an aspiration into a criterion that\nthe v1 tools cannot execute or prove.\n",{"name":5,"description":7,"disable-model-invocation":20},"main","a341ec69437d028e71ea1de51b7f6d9c1238048b","packages\u002Fpi-work\u002Fskills\u002Fwork-authoring","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI","https:\u002F\u002Fwww.centerforairesearch.org\u002F","CAAIR develops applied AI systems that support civil society, public accountability, and democratic infrastructure.","2026-11-12 02:15:25",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add CenterForAgenticAI\u002Ftools --skill work-authoring",{"tool":38,"command":39},"bunx","bunx skills add CenterForAgenticAI\u002Ftools --skill work-authoring",{"tool":41,"command":42},"pnpm","pnpm dlx skills add CenterForAgenticAI\u002Ftools --skill work-authoring","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-work\u002Fskills\u002Fwork-authoring\u002FSKILL.md",true,[],[47,54,61,68,75,82],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M7DBS96C51VE1KHCM1N0SR47","work-execution","Execute a validated pi-work node or wave when you need the ready-set, by-reference work_plan brief, Fabric default, pi-delegate fallback, explicit tree-bound verification, remediation, scope amendments, or a post-integration refactor pass. Use after decomposition and before claiming any node is done.","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-work\u002Fskills\u002Fwork-execution\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M7DBS2SJ4ZND4SG77JX66BWK","work-decomposition","Turn a validated pi-work intent into a strict workspec work graph when the work is too large for one bounded worker. Use when assigning criterion homes, dependencies, honest write scopes, open decisions, or optional model-change nodes before planning execution.","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-work\u002Fskills\u002Fwork-decomposition\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M7DBRST29F8J729DHGCM55WJ","intent-records","Draft or approve an intent record when a person asks for a change, or describe existing behaviour before retrofitting compiled intent. Use before turning intent into a pi-work workspec.","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-intent\u002Fskills\u002Fintent-records\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M7DBRNVNYBW7K25Y7W7CYVEA","intent-conformance","Connect an approved finite Bend model to actual application code with a language-neutral conformance manifest, app-language oracles and a broken-model control. Load when proving implementation parity or diagnosing decision drift.","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-intent\u002Fskills\u002Fintent-conformance\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7DBRH9KT2DMXVT16P1JCHQK","compiled-intent-model","Author or change a finite Bend model from an approved intent record, review its meaning, obtain human law approval, and produce an offline-verifiable Jev receipt. Load for model-change nodes, not for emergent runtime behaviour.","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-intent\u002Fskills\u002Fcompiled-intent-model\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M7DBRBGVD06GXTFWR7GXA67Y","developing-pi-extension-repos","Keep tests, dependencies, CI reproduction, and coverage reliable across Pi extension repositories. Load when editing, testing, or setting up a worktree in a pi extension repository, when tests pass from source but fail from compiled output (or vice versa), before running `npm install` in a checkout other sessions may be using. For writing extension code itself, load `authoring-pi-extensions`.","https:\u002F\u002Fgithub.com\u002FCenterForAgenticAI\u002Ftools\u002Fblob\u002Fmain\u002Fpackages\u002Fpi-dev\u002Fskills\u002Fdeveloping-pi-extension-repos\u002FSKILL.md",[],[]]