[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C4X70X4P7PCPNE5Y0DKN4T":3,"skill-same:01M7C4X70X4P7PCPNE5Y0DKN4T":330},{"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":28,"commitSha":29,"skillDirectory":30,"forksCount":14,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":33,"ownerDescription":6,"updatedAt":34,"install":35,"resources":48},"01M7C4X70X4P7PCPNE5Y0DKN4T","collab-monitor",null,"Native Monitor for addressed collab mail on Claude\u002Fcmux seats, report DONE markers and installed-version watches. Codex\u002FT3 seats: no polling turns; completion notifications wake you; a T3 lead keeps one lead name waiter. NOT for file-integrity auditing.","EtanHey\u002Fgolems","EtanHey","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F93338119?v=4","01M7C4SKC9HJC3B9TAX0D9MZ1E","Etan Heyman","01M7C4SKCT59Z49TTN4WWDDEYD",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: collab-monitor\ndescription: \"Native Monitor for addressed collab mail on Claude\u002Fcmux seats, report DONE markers and installed-version watches. Codex\u002FT3 seats: no polling turns; completion notifications wake you; a T3 lead keeps one lead name waiter. NOT for file-integrity auditing.\"\nversion: 1.4.0\ntype: encoded-preference\nlast-eval-date: 2026-10-12\ncompliance-score: \"20\u002F20 owner lifecycle and state hygiene regressions; 6\u002F6 episode regressions; 19\u002F19 routing checks; no agent-behavior or native Monitor runtime proof\"\n---\n\n# Collab Monitor\n\nUse the native **Monitor** tool when the harness exposes it. Arm before dispatch,\nkeep the returned watch identifier in the lane record, and stop it when its lane\ncloses. Codex and T3 seats never poll collabs with turns; a T3 lead keeps one name waiter (next section),\nwhich also watches its lane collab (§ Lanes with a collab). A detached producer alone does not wake an agent.\n\n## Codex and T3 seats\n\nThe m1-launch-queue collab is now an admission ledger, not a request queue: [agent-routing\u002Freferences\u002Fcross-machine.md](..\u002Fagent-routing\u002Freferences\u002Fcross-machine.md) § 5.\n\n> \"Codex seats must not poll collabs; T3 delegate completions already wake you.\"\n> \"No polling turns. No collab-monitor wakes in Codex\"\n> (Etan 2026-10-18, via orc collab)\n\nWhy: the weekly Codex quota drained in under 24 h, with the Codex \"Collab\nMonitor\" skill at 196 uses on Oct 18, and \"Collab-monitor polling wakes cost\nanother 201M across 729 turns.\" (Etan 2026-10-18, via orc collab)\n\n- A Codex seat never arms collab-monitor, never runs `start`\u002F`follow`\u002F`run`,\n  and never polls a collab file with model turns, in a loop or on a schedule\n  (a lead's one shell name waiter below is the only exception).\n- A T3 seat (any vendor) relies on its parent-owned wake signals: `delegate_task`\n  completion notifications for child work and `watch_pull_request` for PRs. Do\n  not add a second watcher on top of them, apart from the lead name waiter below.\n- A T3 lead keeps exactly one name waiter armed at all times on the leads+orc\n  collab, `orchestrator\u002Fdocs.local\u002Fcollab\u002Forc-inbox.md` (lead rules §1, 2026-10-30),\n  and on each lane collab it leads (§ Lanes with a collab): a background shell loop\n  that exits on the first new `@\u003Chandle>` or `@all-leads` line in the inbox, or the\n  first new line addressed to its listen name in a lane collab. Register it with `t3_background_wait` and re-arm it on every wake. It\n  costs no model turns until a line matches, so it is not a polling turn. Keep\n  your own handle out of your own posts, or it wakes you. Workers and children never arm it.\n- If something genuinely must sit and watch, dispatch one bounded\n  `codex.subagent.mechanical` child per `\u002Fagent-routing` § Routing rules (SSOT);\n  resolve its model with\n  `node scripts\u002Fmodel-roles.mjs codex.subagent.mechanical --use watcher`.\n  It reports once and exits; it never owns the parent's PR watch.\n- Native Monitor below is for Claude seats in a cmux fleet.\n\n## Lanes with a collab\n\n> \"when there is a collab the message shouldn't travel through the chat, but have\n> the monitor on the […] collab damn it\" (Etan 2026-11-02, via orc)\n\n- Every worker brief says `Report to \u003Ccollab path>, not to the lead thread, then end your task.`\n  Workers append their status, DONE and BLOCKED lines to the collab, then end the task.\n- In T3, the task's end is the real wake. Workers send no separate thread message;\n  they end normally (with the report pointer), and the completion notification\n  wakes the lead, whose turn from it can dispatch the next step.\n- The lead keeps ONE live monitor on the collab, for lines addressed to it. A\n  Claude seat in a cmux fleet uses native Monitor (§ Collab tail with a lead\n  filter), re-armed at expiry. A T3 lead (any vendor) extends the one name waiter\n  above to the lane collab, so it adds no second waiter and no polling turns.\n- In T3, that name waiter is for awareness only. A waiter wake can arrive as a\n  background-task turn, and T3 can refuse `delegate_task` from it\n  (`parent_not_active`, reproduced twice on 2026-11-02; once it did dispatch).\n  Don't dispatch from a waiter-woken T3 turn: a next step follows on the\n  completion wake. Task-completion and PR-watch wakes can dispatch. In T3 the\n  completion notification stays the liveness guard and the dispatch wake; the\n  collab waiter carries the messages.\n- On a Claude cmux seat, the native Monitor is the completion watch too (§ Native\n  Monitor first): act on an addressed BLOCKED\u002FDONE line from its wake, including\n  dispatching the next step.\n- Don't relay worker updates into chat; chat is for Etan and orc's routing. No polling turns.\n\n## Native Monitor first\n\nCall `Monitor({command, description, timeout_ms: 1800000})`: a command, a short\ndescription naming the lane, and a 30-minute timeout. This Claude Code harness\ncaps the tool at 1800000 ms; other harness versions may differ, so check their\nschema. Give it events that each require action; use one completion watch for\na batch of tests or CI jobs.\n\nKeep every timer wake inside the prompt-cache window (Claude ~60 min, Codex\n~30 min): a Monitor `timeout_ms`, `ScheduleWakeup` delay or re-arm interval is\n≤50 min (3000000 ms) on a Claude seat and ≤25 min on a Codex seat. Never arm\n3600 s: wakes past the window busted the cache 20 times in a week, costing\nClaude 7.4M tokens (35 Monitor arms at 3600 s, cache audit 2026-10-19). Prefer\nan event wake (a completion notification or `watch_pull_request`) over a timer.\n\n### Collab tail with a lead filter\n\nUse the foreground routing filter as the native tool's command. Monitor owns the\nprocess and wakeups; the filter owns watermarks, hash dedup and author matching.\nThis avoids spawning a detached producer before every native watch.\n\n```text\nMonitor({\n  command: 'bash \"$HOME\u002F.claude\u002Fskills\u002Fcollab-monitor\u002Fscripts\u002Fcollab-monitor.sh\" run --alias @\u003Cseat-id> @\u003Clead-listen-name> \u002Fabsolute\u002Fpath\u002Fcollab.md',\n  description: '\u003Clane> addressed collab mail',\n  timeout_ms: 1800000,\n})\n```\n\nReplace the listen name, seat alias and file with the actual lane values. Use a\nprivate, explicit state root via `MONITOR_STATE_DIR` when isolating a test watch.\nResolve the script from the skill location supplied by your harness; the example\nuses this MacBook's installed `~\u002F.claude\u002Fskills\u002Fcollab-monitor` symlink. Verify\nthe resolved script exists before arming; use its absolute path on other hosts.\nNever point a test watch at a real lane's state root.\n\nThe bounded lead filter accepts word-bounded `@\u003Cname>` mentions, headers addressed\n`### author → name` (also `->`), and that lead's own `\u003Cname>-w\u003CN>` \u002F `\u003Cname>-r\u003CN>`\nDONE\u002FBLOCKED lines. It excludes self-authored blocks, signature-only tags,\nnear-miss names and unrelated headings. Fenced and indented code cannot route\nmail; an unclosed fence reports an incomplete poll and is retried. Add `--alias`\nfor each actual delivery name, not a broad prefix.\n\nThe first poll seeds existing history silently. Re-arming with the same state\nroot preserves dedup. New mail should produce `NEW-FOR-@\u003Cname>`; readiness alone\nis not proof that the consumer received a message. Append collab posts with\n`cat >>` or an append-only writer.\n\nRepeated identical headings in one append-only file are separate occurrences.\nThe first retains its content hash; later ones carry `episode=2`, `episode=3`,\netc. Read that routed occurrence outside code in the alert's file for its report body\nand pointer. Extending a body does not create another heading episode. Restart\npreserves dedup; existing heading-only state silently seeds pre-watermark repeated\nheadings on its next successful growth poll. Identical copies at the same\noccurrence in different watched files still share a dedup key. Use a unique final\ncompletion heading per report episode so older installed monitors also catch it.\nThis repair does not attach a parent consumer or replay a missed report.\n\n### Report-file DONE markers\n\nSet the report path and the exact marker in this command, then pass it as\nMonitor's `command` with `timeout_ms: 1800000`. Check the **exact final line**,\nnot a DONE word in the report body. This watch completes once; read the report\nand verify its requested artifact after it fires.\n\n```bash report-done\nREPORT='\u002Fabsolute\u002Fprivate\u002Freport.md'\nDONE_MARKER='DONE_LANE'\ndeadline=$(( $(date +%s) + 1800 ))\nwhile true; do\n  if [ -r \"$REPORT\" ] && [ \"$(tail -n 1 \"$REPORT\")\" = \"$DONE_MARKER\" ]; then\n    printf 'REPORT_DONE %s\\n' \"$REPORT\"\n    exit 0\n  fi\n  if [ \"$(date +%s)\" -ge \"$deadline\" ]; then\n    printf 'REARM_REQUIRED report watch\\n'\n    exit 0\n  fi\n  sleep 1 # Cheap local file-stat poll cadence; not an urgency signal.\ndone\n```\n\nA marker is a handoff signal, not proof of merge, publication or installation.\nIf the report is missing or malformed, the watch stays open until its deadline.\n\n### Installed-version watches\n\nQuery the installed command, not the checkout, a merge receipt or a release tag.\nSet the exact expected `--version` output and pass this body to native Monitor.\nThe command re-reads the installed binary on every tick, including replacements\nat the same path.\n\n```bash installed-version\nTOOL='\u002Fabsolute\u002Fpath\u002Finstalled-command'\nEXPECTED_VERSION='expected exact version output'\ndeadline=$(( $(date +%s) + 1800 ))\nwhile true; do\n  if current=\"$(\"$TOOL\" --version 2>\u002Fdev\u002Fnull)\" && [ \"$current\" = \"$EXPECTED_VERSION\" ]; then\n    printf 'VERSION_MATCH %s\\n' \"$current\"\n    exit 0\n  fi\n  if [ \"$(date +%s)\" -ge \"$deadline\" ]; then\n    printf 'REARM_REQUIRED installed-version watch\\n'\n    exit 0\n  fi\n  sleep 1 # Cheap local file-stat poll cadence; not an urgency signal.\ndone\n```\n\nAfter a version match, run the required **real-client smoke**. A version string\nalone does not prove the installed service serves requests successfully.\n\n### Re-arm at 30 minutes\n\nThe 30-minute re-arm is the tool's **hard expiry**, not an optional reminder.\nOn expiry and **after every compaction**, re-check which lanes remain open and\nre-arm their watches using current paths, identifiers and expected versions. Report\u002Fversion examples emit `REARM_REQUIRED` after their 30-minute\nbudget. For a collab stream, cancel the prior native watch by its returned\nidentifier before replacing it; keep the same routing state to avoid history\nreplay. In this Claude Code harness, use **TaskStop** with the returned task id\nto cancel early; check the current schema on other harness versions. Do not\nleave duplicate streams attached.\n\nWhen a watch ends unexpectedly, repair it and re-arm before dispatch resumes.\nWhen its task finishes, stop it. Workers do not monitor their lead's reviewer.\nReview routing: `\u002Fagent-routing` § Routing rules (SSOT). Follow an engine-issued\nmailbox contract when one exists.\n\n## Legacy cmux-only adapter\n\nOnly for a Codex seat running inside a cmux fleet with no T3 parent and no\nnative Monitor. Any T3 seat, or any Codex seat that can rely on completion\nnotifications, stops at [Codex and T3 seats](#codex-and-t3-seats). Record\nits listen name and PID, verify readiness, and attach its consumer to the\norchestrator's monitored command session:\n\n```bash\nCM=\"$HOME\u002F.claude\u002Fskills\u002Fcollab-monitor\u002Fscripts\u002Fcollab-monitor.sh\"\n# Or set CM to the absolute script path from your harness's skill catalog.\ntest -f \"$CM\" || { printf 'Missing collab-monitor script: %s\\n' \"$CM\" >&2; exit 1; }\nbash \"$CM\" start --alias @\u003Cseat-id> @\u003Clisten-name> \u002Fabsolute\u002Fpath\u002Fcollab.md\nbash \"$CM\" status @\u003Clisten-name>\nbash \"$CM\" follow @\u003Clisten-name>\n```\n\n`run`, `follow`, and the `start` producer exit cleanly when their owning agent\ndies. Ownership is the nearest `claude`\u002F`codex` ancestor, or the initial parent\nwhen no agent ancestor exists; `--owner-pid \u003Cpid>` overrides it on all three\ncommands. `start` captures ownership before detaching and persists independently\nof the invoking shell while that owner lives. Foreground `run`\u002F`follow` also\nexit on reparenting to PID 1; the packaged detached producer is exempt from that\nparent check. Owner checks continue during sleep\u002Ftail waits (at most one second\napart), and normal cleanup reaps sleep\u002Ftail children and releases owned state.\n`follow` replays the\ncurrent session log then streams new alerts, and exits when that named watch\nstops. It is the attached consumer, not proof of a native Monitor tool on Codex.\nFor a caller-owned foreground process, use `run`; for a scheduled single poll,\nuse `run --once`. Both use the same bounded routing filter.\n\nAfter handoff or lane closure:\n\n```bash\nbash \"$CM\" stop @\u003Clisten-name>\n```\n\n`status` is read-only: it never acquires locks or prunes state. `start` prunes\ninactive runtime records across its state root:\nPID\u002Finstance\u002Fowner\u002Fready records, stale run locks and abandoned event extracts.\nPruning retains logs, dedup hashes and size watermarks. A live producer or foreground\nrun keeps all its files, including when its token is inconsistent; active or\nownerless start locks younger than 30 seconds are preserved; older ownerless\nlocks are recoverable after interrupted publication. Pruning has signal cleanup.\nReused PIDs belonging to unrelated processes\nare pruned without signaling them. `start` records the owner PID and `status`\nprints `owner_pid` (`unknown` for older instances). Instance identity works across\ncheckout paths. `run` and `follow` exit and reap their children if the script or\nany packaged helper disappears, including during long waits.\n\nStop by listen name, never by a process-name pattern. The implementation checks\nPID and per-start identity before signaling. On `STATE_CONFLICT`, inspect the\npreserved records; do not delete state or signal an unverified process.\n\nDefaults: state `~\u002F.local\u002Fstate\u002Fcollab-monitor\u002F\u003Clisten-name>\u002F`, positive\n`POLL_SECONDS`, readiness budget 30 seconds. `START_TIMEOUT_SECONDS` can extend\nreadiness to at most 86400 seconds; malformed intervals and names `.` \u002F `..`\nare rejected. `--include-self` is for explicit audits only and labels those\nrecords `SELF-POST` rather than inbound mail. Never share test state with a lane.\n\n## Limits and supervision\n\nThe filter detects byte-size shrink, retries transient missing\u002Funreadable files,\nand reports incomplete polls. It does not detect same-size rewrites, messages\noutside the routing grammar, process death without a supervisor, or completion\nvisible only in a worker registry. A crash between alert emission and persistence\ncan duplicate the final alert: delivery is at-least-once.\n\nYou MUST NOT poll `read_screen` in a loop for a worker outcome. For headless Codex\nworkers use `codex-workflows`' `watch`: observe process completion, then read its\nfinished output once. A monitor observes an artifact, not whether a reviewer was\nactually routed. Keep handoff ownership explicit in the lane's `closure` record.\n\n## Recurring payloads\n\nUse live queries on every tick; never copy a stale PR state into the payload.\nFrame the tick with a timestamp, cycle and last verified action. Counters reset\nonly on a verified side effect. Finish with dispatch, verify-and-decrement or\nescalate-park. Codex and T3 seats write no recurring collab-reading ticks\n([Codex and T3 seats](#codex-and-t3-seats)). Scheduler-specific fallback guidance is in\n[references\u002Fcron-payloads.md](references\u002Fcron-payloads.md); it is not a prerequisite\nfor native Monitor.\n\n## Evaluation\n\n```bash\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fmonitor-howto.py\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fepisode-dedup.py\nbash skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Frun-evals.sh candidate\nbash skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Flive-two-file-smoke.sh\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fowner-exit.py\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fstate-hygiene.py\n```\n\nThe how-to checks exercise synthetic report\u002Fversion command transitions and\nteaching contracts. The routing suite exercises the packaged filter and detached\nfallback. Neither proves native Monitor execution on a harness where it is\nunavailable. Agent-behavior scenarios are recorded separately in `evals.json`.\n","# Collab Monitor\n\nUse the native **Monitor** tool when the harness exposes it. Arm before dispatch,\nkeep the returned watch identifier in the lane record, and stop it when its lane\ncloses. Codex and T3 seats never poll collabs with turns; a T3 lead keeps one name waiter (next section),\nwhich also watches its lane collab (§ Lanes with a collab). A detached producer alone does not wake an agent.\n\n## Codex and T3 seats\n\nThe m1-launch-queue collab is now an admission ledger, not a request queue: [agent-routing\u002Freferences\u002Fcross-machine.md](..\u002Fagent-routing\u002Freferences\u002Fcross-machine.md) § 5.\n\n> \"Codex seats must not poll collabs; T3 delegate completions already wake you.\"\n> \"No polling turns. No collab-monitor wakes in Codex\"\n> (Etan 2026-10-18, via orc collab)\n\nWhy: the weekly Codex quota drained in under 24 h, with the Codex \"Collab\nMonitor\" skill at 196 uses on Oct 18, and \"Collab-monitor polling wakes cost\nanother 201M across 729 turns.\" (Etan 2026-10-18, via orc collab)\n\n- A Codex seat never arms collab-monitor, never runs `start`\u002F`follow`\u002F`run`,\n  and never polls a collab file with model turns, in a loop or on a schedule\n  (a lead's one shell name waiter below is the only exception).\n- A T3 seat (any vendor) relies on its parent-owned wake signals: `delegate_task`\n  completion notifications for child work and `watch_pull_request` for PRs. Do\n  not add a second watcher on top of them, apart from the lead name waiter below.\n- A T3 lead keeps exactly one name waiter armed at all times on the leads+orc\n  collab, `orchestrator\u002Fdocs.local\u002Fcollab\u002Forc-inbox.md` (lead rules §1, 2026-10-30),\n  and on each lane collab it leads (§ Lanes with a collab): a background shell loop\n  that exits on the first new `@\u003Chandle>` or `@all-leads` line in the inbox, or the\n  first new line addressed to its listen name in a lane collab. Register it with `t3_background_wait` and re-arm it on every wake. It\n  costs no model turns until a line matches, so it is not a polling turn. Keep\n  your own handle out of your own posts, or it wakes you. Workers and children never arm it.\n- If something genuinely must sit and watch, dispatch one bounded\n  `codex.subagent.mechanical` child per `\u002Fagent-routing` § Routing rules (SSOT);\n  resolve its model with\n  `node scripts\u002Fmodel-roles.mjs codex.subagent.mechanical --use watcher`.\n  It reports once and exits; it never owns the parent's PR watch.\n- Native Monitor below is for Claude seats in a cmux fleet.\n\n## Lanes with a collab\n\n> \"when there is a collab the message shouldn't travel through the chat, but have\n> the monitor on the […] collab damn it\" (Etan 2026-11-02, via orc)\n\n- Every worker brief says `Report to \u003Ccollab path>, not to the lead thread, then end your task.`\n  Workers append their status, DONE and BLOCKED lines to the collab, then end the task.\n- In T3, the task's end is the real wake. Workers send no separate thread message;\n  they end normally (with the report pointer), and the completion notification\n  wakes the lead, whose turn from it can dispatch the next step.\n- The lead keeps ONE live monitor on the collab, for lines addressed to it. A\n  Claude seat in a cmux fleet uses native Monitor (§ Collab tail with a lead\n  filter), re-armed at expiry. A T3 lead (any vendor) extends the one name waiter\n  above to the lane collab, so it adds no second waiter and no polling turns.\n- In T3, that name waiter is for awareness only. A waiter wake can arrive as a\n  background-task turn, and T3 can refuse `delegate_task` from it\n  (`parent_not_active`, reproduced twice on 2026-11-02; once it did dispatch).\n  Don't dispatch from a waiter-woken T3 turn: a next step follows on the\n  completion wake. Task-completion and PR-watch wakes can dispatch. In T3 the\n  completion notification stays the liveness guard and the dispatch wake; the\n  collab waiter carries the messages.\n- On a Claude cmux seat, the native Monitor is the completion watch too (§ Native\n  Monitor first): act on an addressed BLOCKED\u002FDONE line from its wake, including\n  dispatching the next step.\n- Don't relay worker updates into chat; chat is for Etan and orc's routing. No polling turns.\n\n## Native Monitor first\n\nCall `Monitor({command, description, timeout_ms: 1800000})`: a command, a short\ndescription naming the lane, and a 30-minute timeout. This Claude Code harness\ncaps the tool at 1800000 ms; other harness versions may differ, so check their\nschema. Give it events that each require action; use one completion watch for\na batch of tests or CI jobs.\n\nKeep every timer wake inside the prompt-cache window (Claude ~60 min, Codex\n~30 min): a Monitor `timeout_ms`, `ScheduleWakeup` delay or re-arm interval is\n≤50 min (3000000 ms) on a Claude seat and ≤25 min on a Codex seat. Never arm\n3600 s: wakes past the window busted the cache 20 times in a week, costing\nClaude 7.4M tokens (35 Monitor arms at 3600 s, cache audit 2026-10-19). Prefer\nan event wake (a completion notification or `watch_pull_request`) over a timer.\n\n### Collab tail with a lead filter\n\nUse the foreground routing filter as the native tool's command. Monitor owns the\nprocess and wakeups; the filter owns watermarks, hash dedup and author matching.\nThis avoids spawning a detached producer before every native watch.\n\n```text\nMonitor({\n  command: 'bash \"$HOME\u002F.claude\u002Fskills\u002Fcollab-monitor\u002Fscripts\u002Fcollab-monitor.sh\" run --alias @\u003Cseat-id> @\u003Clead-listen-name> \u002Fabsolute\u002Fpath\u002Fcollab.md',\n  description: '\u003Clane> addressed collab mail',\n  timeout_ms: 1800000,\n})\n```\n\nReplace the listen name, seat alias and file with the actual lane values. Use a\nprivate, explicit state root via `MONITOR_STATE_DIR` when isolating a test watch.\nResolve the script from the skill location supplied by your harness; the example\nuses this MacBook's installed `~\u002F.claude\u002Fskills\u002Fcollab-monitor` symlink. Verify\nthe resolved script exists before arming; use its absolute path on other hosts.\nNever point a test watch at a real lane's state root.\n\nThe bounded lead filter accepts word-bounded `@\u003Cname>` mentions, headers addressed\n`### author → name` (also `->`), and that lead's own `\u003Cname>-w\u003CN>` \u002F `\u003Cname>-r\u003CN>`\nDONE\u002FBLOCKED lines. It excludes self-authored blocks, signature-only tags,\nnear-miss names and unrelated headings. Fenced and indented code cannot route\nmail; an unclosed fence reports an incomplete poll and is retried. Add `--alias`\nfor each actual delivery name, not a broad prefix.\n\nThe first poll seeds existing history silently. Re-arming with the same state\nroot preserves dedup. New mail should produce `NEW-FOR-@\u003Cname>`; readiness alone\nis not proof that the consumer received a message. Append collab posts with\n`cat >>` or an append-only writer.\n\nRepeated identical headings in one append-only file are separate occurrences.\nThe first retains its content hash; later ones carry `episode=2`, `episode=3`,\netc. Read that routed occurrence outside code in the alert's file for its report body\nand pointer. Extending a body does not create another heading episode. Restart\npreserves dedup; existing heading-only state silently seeds pre-watermark repeated\nheadings on its next successful growth poll. Identical copies at the same\noccurrence in different watched files still share a dedup key. Use a unique final\ncompletion heading per report episode so older installed monitors also catch it.\nThis repair does not attach a parent consumer or replay a missed report.\n\n### Report-file DONE markers\n\nSet the report path and the exact marker in this command, then pass it as\nMonitor's `command` with `timeout_ms: 1800000`. Check the **exact final line**,\nnot a DONE word in the report body. This watch completes once; read the report\nand verify its requested artifact after it fires.\n\n```bash report-done\nREPORT='\u002Fabsolute\u002Fprivate\u002Freport.md'\nDONE_MARKER='DONE_LANE'\ndeadline=$(( $(date +%s) + 1800 ))\nwhile true; do\n  if [ -r \"$REPORT\" ] && [ \"$(tail -n 1 \"$REPORT\")\" = \"$DONE_MARKER\" ]; then\n    printf 'REPORT_DONE %s\\n' \"$REPORT\"\n    exit 0\n  fi\n  if [ \"$(date +%s)\" -ge \"$deadline\" ]; then\n    printf 'REARM_REQUIRED report watch\\n'\n    exit 0\n  fi\n  sleep 1 # Cheap local file-stat poll cadence; not an urgency signal.\ndone\n```\n\nA marker is a handoff signal, not proof of merge, publication or installation.\nIf the report is missing or malformed, the watch stays open until its deadline.\n\n### Installed-version watches\n\nQuery the installed command, not the checkout, a merge receipt or a release tag.\nSet the exact expected `--version` output and pass this body to native Monitor.\nThe command re-reads the installed binary on every tick, including replacements\nat the same path.\n\n```bash installed-version\nTOOL='\u002Fabsolute\u002Fpath\u002Finstalled-command'\nEXPECTED_VERSION='expected exact version output'\ndeadline=$(( $(date +%s) + 1800 ))\nwhile true; do\n  if current=\"$(\"$TOOL\" --version 2>\u002Fdev\u002Fnull)\" && [ \"$current\" = \"$EXPECTED_VERSION\" ]; then\n    printf 'VERSION_MATCH %s\\n' \"$current\"\n    exit 0\n  fi\n  if [ \"$(date +%s)\" -ge \"$deadline\" ]; then\n    printf 'REARM_REQUIRED installed-version watch\\n'\n    exit 0\n  fi\n  sleep 1 # Cheap local file-stat poll cadence; not an urgency signal.\ndone\n```\n\nAfter a version match, run the required **real-client smoke**. A version string\nalone does not prove the installed service serves requests successfully.\n\n### Re-arm at 30 minutes\n\nThe 30-minute re-arm is the tool's **hard expiry**, not an optional reminder.\nOn expiry and **after every compaction**, re-check which lanes remain open and\nre-arm their watches using current paths, identifiers and expected versions. Report\u002Fversion examples emit `REARM_REQUIRED` after their 30-minute\nbudget. For a collab stream, cancel the prior native watch by its returned\nidentifier before replacing it; keep the same routing state to avoid history\nreplay. In this Claude Code harness, use **TaskStop** with the returned task id\nto cancel early; check the current schema on other harness versions. Do not\nleave duplicate streams attached.\n\nWhen a watch ends unexpectedly, repair it and re-arm before dispatch resumes.\nWhen its task finishes, stop it. Workers do not monitor their lead's reviewer.\nReview routing: `\u002Fagent-routing` § Routing rules (SSOT). Follow an engine-issued\nmailbox contract when one exists.\n\n## Legacy cmux-only adapter\n\nOnly for a Codex seat running inside a cmux fleet with no T3 parent and no\nnative Monitor. Any T3 seat, or any Codex seat that can rely on completion\nnotifications, stops at [Codex and T3 seats](#codex-and-t3-seats). Record\nits listen name and PID, verify readiness, and attach its consumer to the\norchestrator's monitored command session:\n\n```bash\nCM=\"$HOME\u002F.claude\u002Fskills\u002Fcollab-monitor\u002Fscripts\u002Fcollab-monitor.sh\"\n# Or set CM to the absolute script path from your harness's skill catalog.\ntest -f \"$CM\" || { printf 'Missing collab-monitor script: %s\\n' \"$CM\" >&2; exit 1; }\nbash \"$CM\" start --alias @\u003Cseat-id> @\u003Clisten-name> \u002Fabsolute\u002Fpath\u002Fcollab.md\nbash \"$CM\" status @\u003Clisten-name>\nbash \"$CM\" follow @\u003Clisten-name>\n```\n\n`run`, `follow`, and the `start` producer exit cleanly when their owning agent\ndies. Ownership is the nearest `claude`\u002F`codex` ancestor, or the initial parent\nwhen no agent ancestor exists; `--owner-pid \u003Cpid>` overrides it on all three\ncommands. `start` captures ownership before detaching and persists independently\nof the invoking shell while that owner lives. Foreground `run`\u002F`follow` also\nexit on reparenting to PID 1; the packaged detached producer is exempt from that\nparent check. Owner checks continue during sleep\u002Ftail waits (at most one second\napart), and normal cleanup reaps sleep\u002Ftail children and releases owned state.\n`follow` replays the\ncurrent session log then streams new alerts, and exits when that named watch\nstops. It is the attached consumer, not proof of a native Monitor tool on Codex.\nFor a caller-owned foreground process, use `run`; for a scheduled single poll,\nuse `run --once`. Both use the same bounded routing filter.\n\nAfter handoff or lane closure:\n\n```bash\nbash \"$CM\" stop @\u003Clisten-name>\n```\n\n`status` is read-only: it never acquires locks or prunes state. `start` prunes\ninactive runtime records across its state root:\nPID\u002Finstance\u002Fowner\u002Fready records, stale run locks and abandoned event extracts.\nPruning retains logs, dedup hashes and size watermarks. A live producer or foreground\nrun keeps all its files, including when its token is inconsistent; active or\nownerless start locks younger than 30 seconds are preserved; older ownerless\nlocks are recoverable after interrupted publication. Pruning has signal cleanup.\nReused PIDs belonging to unrelated processes\nare pruned without signaling them. `start` records the owner PID and `status`\nprints `owner_pid` (`unknown` for older instances). Instance identity works across\ncheckout paths. `run` and `follow` exit and reap their children if the script or\nany packaged helper disappears, including during long waits.\n\nStop by listen name, never by a process-name pattern. The implementation checks\nPID and per-start identity before signaling. On `STATE_CONFLICT`, inspect the\npreserved records; do not delete state or signal an unverified process.\n\nDefaults: state `~\u002F.local\u002Fstate\u002Fcollab-monitor\u002F\u003Clisten-name>\u002F`, positive\n`POLL_SECONDS`, readiness budget 30 seconds. `START_TIMEOUT_SECONDS` can extend\nreadiness to at most 86400 seconds; malformed intervals and names `.` \u002F `..`\nare rejected. `--include-self` is for explicit audits only and labels those\nrecords `SELF-POST` rather than inbound mail. Never share test state with a lane.\n\n## Limits and supervision\n\nThe filter detects byte-size shrink, retries transient missing\u002Funreadable files,\nand reports incomplete polls. It does not detect same-size rewrites, messages\noutside the routing grammar, process death without a supervisor, or completion\nvisible only in a worker registry. A crash between alert emission and persistence\ncan duplicate the final alert: delivery is at-least-once.\n\nYou MUST NOT poll `read_screen` in a loop for a worker outcome. For headless Codex\nworkers use `codex-workflows`' `watch`: observe process completion, then read its\nfinished output once. A monitor observes an artifact, not whether a reviewer was\nactually routed. Keep handoff ownership explicit in the lane's `closure` record.\n\n## Recurring payloads\n\nUse live queries on every tick; never copy a stale PR state into the payload.\nFrame the tick with a timestamp, cycle and last verified action. Counters reset\nonly on a verified side effect. Finish with dispatch, verify-and-decrement or\nescalate-park. Codex and T3 seats write no recurring collab-reading ticks\n([Codex and T3 seats](#codex-and-t3-seats)). Scheduler-specific fallback guidance is in\n[references\u002Fcron-payloads.md](references\u002Fcron-payloads.md); it is not a prerequisite\nfor native Monitor.\n\n## Evaluation\n\n```bash\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fmonitor-howto.py\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fepisode-dedup.py\nbash skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Frun-evals.sh candidate\nbash skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Flive-two-file-smoke.sh\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fowner-exit.py\npython3 skills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fstate-hygiene.py\n```\n\nThe how-to checks exercise synthetic report\u002Fversion command transitions and\nteaching contracts. The routing suite exercises the packaged filter and detached\nfallback. Neither proves native Monitor execution on a harness where it is\nunavailable. Agent-behavior scenarios are recorded separately in `evals.json`.\n",{"name":5,"type":24,"version":25,"description":7,"last-eval-date":26,"compliance-score":27},"encoded-preference","1.4.0",1791763200000,"20\u002F20 owner lifecycle and state hygiene regressions; 6\u002F6 episode regressions; 19\u002F19 routing checks; no agent-behavior or native Monitor runtime proof","master","3400bc971628f00ab7bf1063993477e211a5508a","skills\u002Fgolem-powers\u002Fcollab-monitor","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems","https:\u002F\u002Fgithub.com\u002FEtanHey","etanheyman.com","2026-11-15 03:05:08",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add EtanHey\u002Fgolems --skill collab-monitor",{"tool":41,"command":42},"bunx","bunx skills add EtanHey\u002Fgolems --skill collab-monitor",{"tool":44,"command":45},"pnpm","pnpm dlx skills add EtanHey\u002Fgolems --skill collab-monitor","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002FSKILL.md",true,[49,55,61,65,69,73,77,81,85,89,93,97,101,105,109,113,117,121,125,129,133,137,141,145,149,153,157,161,165,169,173,177,181,185,189,192,196,200,204,208,212,216,220,224,228,232,236,240,244,248,252,256,260,266,270,274,278,282,286,290,294,298,302,306,310,314,318,322,326],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"evals\u002Fcron-payloads-red-vs-green.md","REFERENCE","参考文档",811,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fcron-payloads-red-vs-green.md",{"path":56,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":59,"sourceHtmlUrl":60},"evals\u002Fdetached-launch.sh","SCRIPT","脚本",1251,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fdetached-launch.sh",{"path":62,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":63,"sourceHtmlUrl":64},"evals\u002Fepisode-dedup.py",7245,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fepisode-dedup.py",{"path":66,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":67,"sourceHtmlUrl":68},"evals\u002Fevals.json",38964,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fevals.json",{"path":70,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":71,"sourceHtmlUrl":72},"evals\u002Ffixtures\u002Faddressed-event-two.md",94,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Faddressed-event-two.md",{"path":74,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":75,"sourceHtmlUrl":76},"evals\u002Ffixtures\u002Faddressed-event.md",122,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Faddressed-event.md",{"path":78,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":79,"sourceHtmlUrl":80},"evals\u002Ffixtures\u002Farrowed-authors.md",490,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Farrowed-authors.md",{"path":82,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":83,"sourceHtmlUrl":84},"evals\u002Ffixtures\u002Fcron-payloads\u002Fcompliant-drive-to-completion-tick.md",619,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fcron-payloads\u002Fcompliant-drive-to-completion-tick.md",{"path":86,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":87,"sourceHtmlUrl":88},"evals\u002Ffixtures\u002Fcron-payloads\u002Fcompliant-pr-gate-payload.md",557,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fcron-payloads\u002Fcompliant-pr-gate-payload.md",{"path":90,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":91,"sourceHtmlUrl":92},"evals\u002Ffixtures\u002Fcron-payloads\u002Fcompliant-rotation-plan.md",544,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fcron-payloads\u002Fcompliant-rotation-plan.md",{"path":94,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":95,"sourceHtmlUrl":96},"evals\u002Ffixtures\u002Fcron-payloads\u002Fstale-pr-gate-payload.md",495,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fcron-payloads\u002Fstale-pr-gate-payload.md",{"path":98,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":99,"sourceHtmlUrl":100},"evals\u002Ffixtures\u002Fdefault-author-filter.md",367,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fdefault-author-filter.md",{"path":102,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":103,"sourceHtmlUrl":104},"evals\u002Ffixtures\u002Fdirect-event-under-mention-heading.md",80,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fdirect-event-under-mention-heading.md",{"path":106,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":107,"sourceHtmlUrl":108},"evals\u002Ffixtures\u002Fepisode-dedup.json",1315,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fepisode-dedup.json",{"path":110,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":111,"sourceHtmlUrl":112},"evals\u002Ffixtures\u002Ffake-failing-event-shasum.sh",150,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ffake-failing-event-shasum.sh",{"path":114,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":115,"sourceHtmlUrl":116},"evals\u002Ffixtures\u002Ffake-failing-seen-mv.sh",147,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ffake-failing-seen-mv.sh",{"path":118,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":119,"sourceHtmlUrl":120},"evals\u002Ffixtures\u002Ffake-failing-wc.sh",175,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ffake-failing-wc.sh",{"path":122,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":123,"sourceHtmlUrl":124},"evals\u002Ffixtures\u002Ffake-slow-shasum.sh",118,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ffake-slow-shasum.sh",{"path":126,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":127,"sourceHtmlUrl":128},"evals\u002Ffixtures\u002Ffake-zombie-ps.sh",192,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ffake-zombie-ps.sh",{"path":130,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":131,"sourceHtmlUrl":132},"evals\u002Ffixtures\u002Ffilter-too-broad.md",162,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ffilter-too-broad.md",{"path":134,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":135,"sourceHtmlUrl":136},"evals\u002Ffixtures\u002Fforeign-event-in-self-block.md",134,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fforeign-event-in-self-block.md",{"path":138,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":139,"sourceHtmlUrl":140},"evals\u002Ffixtures\u002Fforeign-signature-with-cc.md",78,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fforeign-signature-with-cc.md",{"path":142,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":143,"sourceHtmlUrl":144},"evals\u002Ffixtures\u002Flarge-plan-unrelated-append.md",20,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Flarge-plan-unrelated-append.md",{"path":146,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":147,"sourceHtmlUrl":148},"evals\u002Ffixtures\u002Forchestrator-fourteen-posts.md",1271,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Forchestrator-fourteen-posts.md",{"path":150,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":151,"sourceHtmlUrl":152},"evals\u002Ffixtures\u002Fprefix-listener-event.md",56,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fprefix-listener-event.md",{"path":154,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":155,"sourceHtmlUrl":156},"evals\u002Ffixtures\u002Froute-contract.json",28508,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Froute-contract.json",{"path":158,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":159,"sourceHtmlUrl":160},"evals\u002Ffixtures\u002Frouting-backtick-inline-span.md",124,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-backtick-inline-span.md",{"path":162,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":163,"sourceHtmlUrl":164},"evals\u002Ffixtures\u002Frouting-code-examples.md",120,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-code-examples.md",{"path":166,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":167,"sourceHtmlUrl":168},"evals\u002Ffixtures\u002Frouting-compact-list-fence.md",90,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-compact-list-fence.md",{"path":170,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":171,"sourceHtmlUrl":172},"evals\u002Ffixtures\u002Frouting-embedded-recipient.md",74,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-embedded-recipient.md",{"path":174,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":175,"sourceHtmlUrl":176},"evals\u002Ffixtures\u002Frouting-invalid-fence-closer.md",79,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-invalid-fence-closer.md",{"path":178,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":179,"sourceHtmlUrl":180},"evals\u002Ffixtures\u002Frouting-list-nested-fence.md",97,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-list-nested-fence.md",{"path":182,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":183,"sourceHtmlUrl":184},"evals\u002Ffixtures\u002Frouting-list-reply.md",61,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-list-reply.md",{"path":186,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":187,"sourceHtmlUrl":188},"evals\u002Ffixtures\u002Frouting-summary-mention.md",58,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-summary-mention.md",{"path":190,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":151,"sourceHtmlUrl":191},"evals\u002Ffixtures\u002Frouting-unclosed-fence.md","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Frouting-unclosed-fence.md",{"path":193,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":194,"sourceHtmlUrl":195},"evals\u002Ffixtures\u002Fself-authored-post.md",413,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fself-authored-post.md",{"path":197,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":198,"sourceHtmlUrl":199},"evals\u002Ffixtures\u002Fself-signature-body.md",65,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fself-signature-body.md",{"path":201,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":202,"sourceHtmlUrl":203},"evals\u002Ffixtures\u002Fself-signature-line.md",18,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fself-signature-line.md",{"path":205,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":206,"sourceHtmlUrl":207},"evals\u002Ffixtures\u002Fself-signature-with-context.md",141,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fself-signature-with-context.md",{"path":209,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":210,"sourceHtmlUrl":211},"evals\u002Ffixtures\u002Fsupervisor-read-screen-loop.txt",277,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fsupervisor-read-screen-loop.txt",{"path":213,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":214,"sourceHtmlUrl":215},"evals\u002Ffixtures\u002Fsupervisor-watch-once.txt",126,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Fsupervisor-watch-once.txt",{"path":217,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":218,"sourceHtmlUrl":219},"evals\u002Ffixtures\u002Ften-posts-three-fires.md",948,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Ffixtures\u002Ften-posts-three-fires.md",{"path":221,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":222,"sourceHtmlUrl":223},"evals\u002Flarge-plan-teaching.py",11504,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Flarge-plan-teaching.py",{"path":225,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":226,"sourceHtmlUrl":227},"evals\u002Flive-two-file-smoke.sh",1583,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Flive-two-file-smoke.sh",{"path":229,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":230,"sourceHtmlUrl":231},"evals\u002Fmonitor-howto.py",6277,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fmonitor-howto.py",{"path":233,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":234,"sourceHtmlUrl":235},"evals\u002Fowner-exit.py",8695,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fowner-exit.py",{"path":237,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":238,"sourceHtmlUrl":239},"evals\u002Fresults\u002F2026-08-03-deterministic-red-green.json",1907,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002F2026-08-03-deterministic-red-green.json",{"path":241,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":242,"sourceHtmlUrl":243},"evals\u002Fresults\u002F2026-10-05-native-howto-deterministic.json",722,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002F2026-10-05-native-howto-deterministic.json",{"path":245,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":246,"sourceHtmlUrl":247},"evals\u002Fresults\u002F2026-10-12-state-hygiene.json",2493,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002F2026-10-12-state-hygiene.json",{"path":249,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":250,"sourceHtmlUrl":251},"evals\u002Fresults\u002Fcron-payloads-baseline-2026-05-02.json",945,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002Fcron-payloads-baseline-2026-05-02.json",{"path":253,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":254,"sourceHtmlUrl":255},"evals\u002Fresults\u002Fcron-payloads-delta-2026-05-02.json",844,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002Fcron-payloads-delta-2026-05-02.json",{"path":257,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":258,"sourceHtmlUrl":259},"evals\u002Fresults\u002Fcron-payloads-with-skill-2026-05-02.json",956,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002Fcron-payloads-with-skill-2026-05-02.json",{"path":261,"resourceType":262,"resourceTypeDesc":263,"sizeBytes":264,"sourceHtmlUrl":265},"evals\u002Fresults\u002Fgreen-with-skill.log","OTHER","其他",741,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002Fgreen-with-skill.log",{"path":267,"resourceType":262,"resourceTypeDesc":263,"sizeBytes":268,"sourceHtmlUrl":269},"evals\u002Fresults\u002Flive-two-file-smoke.log",193,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002Flive-two-file-smoke.log",{"path":271,"resourceType":262,"resourceTypeDesc":263,"sizeBytes":272,"sourceHtmlUrl":273},"evals\u002Fresults\u002Fred-baseline.log",1215,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fresults\u002Fred-baseline.log",{"path":275,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":276,"sourceHtmlUrl":277},"evals\u002Froute-contract.py",6020,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Froute-contract.py",{"path":279,"resourceType":262,"resourceTypeDesc":263,"sizeBytes":280,"sourceHtmlUrl":281},"evals\u002Frun-evals-parts\u002Fcandidate-01.bash",41676,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Frun-evals-parts\u002Fcandidate-01.bash",{"path":283,"resourceType":262,"resourceTypeDesc":263,"sizeBytes":284,"sourceHtmlUrl":285},"evals\u002Frun-evals-parts\u002Fcandidate-02.bash",8158,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Frun-evals-parts\u002Fcandidate-02.bash",{"path":287,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":288,"sourceHtmlUrl":289},"evals\u002Frun-evals.sh",6507,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Frun-evals.sh",{"path":291,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":292,"sourceHtmlUrl":293},"evals\u002Fstate-hygiene.py",10219,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fevals\u002Fstate-hygiene.py",{"path":295,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":296,"sourceHtmlUrl":297},"references\u002Fcron-payloads.md",10845,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Freferences\u002Fcron-payloads.md",{"path":299,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":300,"sourceHtmlUrl":301},"scripts\u002Fcollab-monitor.sh",2042,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fcollab-monitor.sh",{"path":303,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":304,"sourceHtmlUrl":305},"scripts\u002Fmonitor\u002Flifecycle.sh",9288,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Flifecycle.sh",{"path":307,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":308,"sourceHtmlUrl":309},"scripts\u002Fmonitor\u002Flocks.sh",7597,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Flocks.sh",{"path":311,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":312,"sourceHtmlUrl":313},"scripts\u002Fmonitor\u002Foptions.sh",1840,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Foptions.sh",{"path":315,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":316,"sourceHtmlUrl":317},"scripts\u002Fmonitor\u002Fowner.sh",2066,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Fowner.sh",{"path":319,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":320,"sourceHtmlUrl":321},"scripts\u002Fmonitor\u002Frouting.awk",12528,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Frouting.awk",{"path":323,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":324,"sourceHtmlUrl":325},"scripts\u002Fmonitor\u002Frun.sh",5696,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Frun.sh",{"path":327,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":328,"sourceHtmlUrl":329},"scripts\u002Fmonitor\u002Fstate.sh",8068,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fcollab-monitor\u002Fscripts\u002Fmonitor\u002Fstate.sh",[331,338,345,352,359,366],{"skillNo":332,"name":333,"nameZh":6,"description":334,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":335,"categories":336,"keywords":337,"publishedAt":19},"01M7C53JEF6A6SS20Q0D2YJYM8","yash","Push dependency bugs upstream. Triggers: dependency bug, workaround, monkey-patch, fork\u002Ffix. NOT app bugs.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fyash\u002FSKILL.md",[],[],{"skillNo":339,"name":340,"nameZh":6,"description":341,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":342,"categories":343,"keywords":344,"publishedAt":19},"01M7C53EPMNSZKHK47VY7WQDP0","whats-new","Diff Claude\u002FCodex\u002FCursor\u002FWispr\u002Fagy changelogs vs local setup. Triggers: whats new, changelog, version bump.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fwhats-new\u002FSKILL.md",[],[],{"skillNo":346,"name":347,"nameZh":6,"description":348,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":349,"categories":350,"keywords":351,"publishedAt":19},"01M7C53C4TVQ9G7AB3X57DACPB","weave","Orchestrator convergence: mine JSONLs, cite findings, route actions. Triggers: weave, run weave.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fweave\u002FSKILL.md",[],[],{"skillNo":353,"name":354,"nameZh":6,"description":355,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":356,"categories":357,"keywords":358,"publishedAt":19},"01M7C5342EDDN7YH53KKW511V2","unslop","Cut AI tells while keeping every receipt, number, path, and hedge intact. Triggers: unslop, cut the slop, wall of text, too verbose, rambly, make this readable. NOT for delivery cadence (see i-have-adhd), channel choice, or artifact length governance.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Funslop\u002FSKILL.md",[],[],{"skillNo":360,"name":361,"nameZh":6,"description":362,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":363,"categories":364,"keywords":365,"publishedAt":19},"01M7C52M0CTDW5DDP4452FJV0G","tmp-block","Guard denying durable writes to \u002Ftmp, \u002Fprivate\u002Ftmp, \u002Fvar\u002Ffolders, $TMPDIR; enforces \u003Crepo>\u002F.worktrees\u002F\u003Cname>. Triggers: tmp guard, TMP-BLOCK deny, WORKTREE-CONVENTION deny, bypass ledger. NOT for storage policy.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Ftmp-block\u002FSKILL.md",[],[],{"skillNo":367,"name":368,"nameZh":6,"description":369,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":370,"categories":371,"keywords":372,"publishedAt":19},"01M7C52HN4Q9E1Z74Z11K23E2H","t3-visuals","Show\u002Fvisualize\u002Fchart\u002Fdiagram\u002Fcompare\u002Fstatus board in T3 with compact interactive HTML. Not standalone dashboards.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Ft3-visuals\u002FSKILL.md",[],[]]