[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M758K67TCK51B67HJCSRRAKQ":3,"skill-same:01M758K67TCK51B67HJCSRRAKQ":61},{"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":38,"commitSha":39,"skillDirectory":40,"forksCount":41,"repositoryHtmlUrl":42,"repositoryHomepageUrl":43,"ownerHtmlUrl":44,"ownerBlogUrl":6,"ownerDescription":45,"updatedAt":46,"install":47,"resources":60},"01M758K67TCK51B67HJCSRRAKQ","upstream-fix-watchdog",null,"Watch upstream for a bug fix to land; silent until fixed.","lukemcqueen\u002Fsteadfaste-fleet","lukemcqueen","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F405240?v=4","01M757MHPAGRXTJAT46SKZP98R","Luke McQueen","01M757MHPZ573KP8DT9GB8SBQ7",4,"MIT","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fupstream-fix-watchdog\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: upstream-fix-watchdog\nversion: 1.0.0\ndescription: Watch upstream for a bug fix to land; silent until fixed.\ntags: [cron, watchdog, upstream, fix, fleet, pin, monitoring]\nrelated_skills: [watchers, fleet-commands, cron-format-standard, steadfaste-fleet-maintenance]\n---\n\n# Upstream Fix Watchdog\n\nWhen an upstream project (e.g. hermes-agent by NousResearch) ships a bug that\nforces you to **pin a version and pause fleet update crons**, you need a cheap\nwatchdog that tells you the moment the fix lands — without nagging every day\nwhile it's still broken.\n\nPattern proven 2026-08-25→28: `_stdio_children_dead()` inversion in\nhermes-agent's `tools\u002Fmcp_tool.py` broke every stdio MCP call on all hosts.\nFleet pinned, paused `agent-hermes-update`, and a daily cron\n(`check-hermes-upstream-fix`) watched upstream until the fix appeared, then\nalerted with the exact fleet-action steps. Generalize this for every future\nupstream pin.\n\n## When to use\n\n- Upstream bug forces a pin\u002Fpause; you want to know when it's safe to unpin\n- A dependency you don't control may silently keep a known-bad state\n- You need a \"wake me up the day this changes\" watch on a raw upstream file\n- User asks for \"a watcher for an upstream fix\" \u002F \"notify me when X is fixed\"\n\n## Core design (all non-negotiable)\n\n1. **Silent while bug present** — exit 0 with EMPTY stdout. `no_agent=true`\n   cron watchdog pattern: empty stdout = silent tick, no delivery, no tokens.\n2. **Notify exactly once** — marker file dedupes:\n   `MARKER = ~\u002F.hermes\u002Fstate\u002F\u003Cincident>-notified`; write it after first\n   notify; skip future notifies if it exists.\n3. **Fetch failure = LOUD error** — network\u002FHTTP errors print\n   `CHECK FAILED: ...` to stderr and exit 1, so a broken check can never go\n   silent. A silent watchdog that stopped polling is worse than no watchdog.\n4. **Conservative detection** — when the upstream file's structure changes so\n   the marker can't be parsed, treat as \"not fixed\" (return False), never as\n   \"fixed\". False-negative → one more day of silence; false-positive → wakes\n   the fleet for nothing.\n5. **Concrete buggy-marker anchor** — match the exact inverted line (e.g.\n   `return True  # alive`), PLUS positive proof of the fix (e.g. the\n   corrected `return False` path present, or the function deleted entirely).\n\n## Script skeleton\n\n```python\n#!\u002Fusr\u002Fbin\u002Fenv python3\n\"\"\"Watchdog: notify when upstream \u003Crepo> fixes \u003Cbug>. Silent until then.\"\"\"\nimport os, re, sys, urllib.request\n\nDEFAULT_URL = \"https:\u002F\u002Fraw.githubusercontent.com\u002F\u003Corg>\u002F\u003Crepo>\u002Fmain\u002F\u003Cpath>\"\nMARKER = os.path.expanduser(\"~\u002F.hermes\u002Fstate\u002F\u003Cincident>-notified\")\nBUGGY_MARKER = \"\u003Cexact buggy line>\"\nNOTIFY_TEXT = \"\"\"🟢 UPSTREAM FIX DETECTED: \u003Cbug> is fixed on origin\u002Fmain.\nFleet action:\n  1. \u003Cunpin\u002Fupdate steps>\n  2. \u003Cresume paused crons: hermes cron resume \u003Cjob_id>>\n  3. \u003Cverify: grep -c '\u003Cbuggy-symbol>' \u003Cfile> on each host>\n  4. \u003Cremove the pin once all hosts verify clean>\"\"\"\n\ndef fetch(url: str) -> str:\n    req = urllib.request.Request(url, headers={\"User-Agent\": \"steadfaste-fleet-fix-watchdog\"})\n    with urllib.request.urlopen(req, timeout=30) as resp:\n        return resp.read().decode(\"utf-8\", errors=\"replace\")\n\ndef is_fixed(src: str) -> bool:\n    # 1. Function\u002Fsymbol absent entirely -> bug gone\n    # 2. Extract the function body; if BUGGY_MARKER in body -> not fixed\n    # 3. Else require positive proof (correct return path present)\n    raise NotImplementedError  # per-incident logic\n\ndef main(argv):\n    url = argv[0] if argv and argv[0].startswith(\"http\") else DEFAULT_URL\n    dry = \"--dry-run\" in argv\n    try:\n        src = fetch(url)\n    except Exception as exc:\n        print(f\"CHECK FAILED: could not fetch {url}: {exc}\", file=sys.stderr)\n        return 1\n    if not is_fixed(src):\n        return 0  # silent\n    if not dry and os.path.exists(MARKER):\n        return 0  # already notified\n    print(NOTIFY_TEXT)\n    if not dry:\n        os.makedirs(os.path.dirname(MARKER), exist_ok=True)\n        open(MARKER, \"w\").write(\"notified\\n\")\n    return 0\n\nif __name__ == \"__main__\":\n    sys.exit(main(sys.argv[1:]))\n```\n\nFull working example: `~\u002F.hermes\u002Fscripts\u002Fcheck-hermes-upstream-fix.py`\n(2026-08-25 incident — real `is_fixed()` regex, marker handling, notify text).\n\n## Cron wiring\n\n```\nhermes cron create --name \u003Cname> --schedule '0 9 * * *' \\\n  --script \u003Cname>.py --no-agent --deliver origin\n```\n\n- `no_agent=true` — script stdout IS the delivery; empty = silent tick\n- Daily 09:00 is a sane cadence for upstream fixes (they don't land hourly)\n- Test before enabling: `python3 \u003Cscript>.py --dry-run` (must print the\n  notify text but NOT write the marker), then `python3 \u003Cscript>.py` once with\n  the marker deleted to see a real tick\n- When the fix fires and the fleet converges, **pause the cron**\n  (`cronjob(action='pause')`) — the job is done; don't let it tick\n  pointlessly forever. Keep the script for the next incident.\n\n## When the fix fires — fleet convergence playbook\n\n1. **Verify on the authoritative source first** — `git fetch origin && git\n   log --oneline -1 origin\u002Fmain` or curl the raw file and eyeball the diff.\n   The watchdog is a tripwire, not proof.\n2. **Unpin \u002F update on the orchestrator host first** (esther\u002Fmoses), verify\n   the service actually works (not just \"command exited 0\").\n3. **Dispatch the fleet — ALWAYS, without asking** (Luke directive\n   2026-08-28: *\"always send for these types of updates. you never have to\n   ask and it just wastes time actually\"*). Do NOT pause to ask permission\n   for a fix\u002Fupdate dispatch — it is the default, not a decision point.\n   `orch-bus-fleet-dispatch.py --execute` (sync stale local\n   agent-registry from the bus host first — see `fleet-commands` skill), or\n   `fleet send \u003Cagent> UPDATE_REQUEST ...`.\n   Include ALL agents; the dispatch tool skips dev-agents — send those\n   manually. When the registry's `health_url` is a stale placeholder\n   (e.g. `your-domain.com`), `fleet send` liveness refuses — use `--force`\n   (the bus is the authoritative proof; verify against `bus.archives`).\n   Verify UPDATE_RESULT `git_sha_after` on the AUTHORITATIVE bus\n   (never the local mirror — it lags).\n4. **Resume paused crons** — `hermes cron resume \u003Cjob_id>` for each\n   `agent-hermes-update`-class job paused during the incident.\n5. **Verify per host** — `grep -c '\u003Cbuggy-symbol>' \u003Cfile>` = 0, or confirm\n   the fixed logic.\n6. **Pause the watchdog** (see Cron wiring above).\n\n## Upstream update may ask \"restore local changes now?\" — decide by content\n\nDuring `hermes update` the updater autostashes local changes and asks whether\nto restore. Decide by content, not by default:\n\n- `git stash show --stat stash@{0}` — see what's inside (working tree is\n  clean when the stash exists, so status won't help)\n- **Local workarounds are DELIBERATE here** (fleet preference): lean skill\n  index, cost guard, max-cost preflight. Restore them — they're your\n  environment protection, not accidental edits\n- Check upstream first: `git show HEAD:\u003Cfile> | grep -c \u003Csymbol>` — if\n  upstream now HAS the functionality (e.g. a `lean` mode), the local patch\n  may be partially redundant; restore anyway (semantics may differ)\n- Conflict check: `git rev-list --count \u003Cstash-base>..HEAD -- \u003Cfiles>` — low\n  count = safe; `git stash show -p stash@{0}` inspects WITHOUT mutating\n- ⚠️ **`git stash apply` MUTATES the working tree** — it is NOT a dry run.\n  To inspect: `git stash show -p stash@{0}`. Only `apply`\u002F`pop` when you\n  intend to restore. (Learned 2026-08-28: an \"inspection\" apply created a\n  real conflict mid-update; restored via\n  `git restore --source=HEAD --staged --worktree \u003Cfiles>` + `git rm --cached`\n  for untracked stash files.)\n\n## Pitfalls\n\n- **The upstream \"fix\" may be incomplete — grep sibling call sites (2026-08-28).**\n  The `_stdio_children_dead()` inversion was fixed upstream, but the same\n  feature's `_watch_stdio_children` probe still had its own coroutine-leak\n  bug: `inspect.isawaitable(_watch_children())` INVOKES the async function,\n  creating a never-awaited coroutine per MCP call (`RuntimeWarning: coroutine\n  'MCPServerTask._watch_stdio_children' was never awaited`). On Titus this\n  made loop-governance MCP calls fail silently — `begin_change` never\n  completed, so the enforcer blocked ALL write tools \"even for inspection.\"\n  **Before trusting an upstream fix, grep the whole feature for the same\n  flaw class** — a fix to one function often leaves siblings broken. See\n  `references\u002Fmcp-tool-watch-probe-case.md` for the full reproduction.\n- **Seam-rot on the LOCAL patch re-apply (whole-deploy abort).** Steadfaste Fleet\n  re-applies local hermes-agent patches after every `hermes update` via\n  `ops\u002Fscripts\u002Finstall\u002Finstall-*.py` (lean index, cron cost-tracking, mcp-tool\n  watch). When upstream refactors the seam, the installer's marker anchor no\n  longer matches → it prints `FAIL ... anchor not found` and exits 2, which\n  aborts `fleet-update.sh` under `set -euo pipefail` BEFORE the state-save\n  writes `state\u002Fupdate-commit`. Net: the deploy looks normal but the doctor's\n  `Deploy sync` stays red forever. Diagnose by mechanism (`fleet-update.sh`\n  exit 0 is NOT clean — grep the log for `[\u003Cinstaller>] FAIL ... anchor not\n  found`, compare `state\u002Fupdate-commit` to `git rev-parse HEAD`). Distinguish\n  load-bearing (re-target the anchor — e.g. keeping `coding_context: lean`\n  distinct when upstream folds it into `focus`) from superseded (mechanism\n  removed upstream — convert to treat \"absent upstream\" as success, not\n  drift-fail), then re-target in REPO source, verify markers match, COMMIT (a\n  concurrent workflow can `git reset` and sweep uncommitted edits), then\n  deploy. Verify behavior with a live import (`_coding_mode({'coding_context':\n  'lean'})` must return `lean`), not just exit 0.\n- **Marker file must be per-incident** — reuse the same `\u003Cname>` for a new\n  incident and the old marker suppresses the new notify. Name it after the\n  incident (`upstream-hermes-fix-notified`), not generically.\n- **Raw URL vs git clone** — `raw.githubusercontent.com` is the cheap fetch\n  (single file, no auth, no clone). Only clone when you need the whole tree.\n- **Don't trust `return True`\u002F`False` alone** — the exact inverted line is\n  the anchor; the fix's positive return path is the proof. Stubs that remove\n  all returns are NOT fixes.\n- **Delivery target `origin`** — one alert to the operator's DM is the\n  point; don't fan out to every agent.\n- **Keep the script after the incident** — the next upstream break is a\n  5-minute edit (URL + marker + notify text), not a new build.\n- **`set -euo pipefail` assignment trap** — if the script is bash and reads\n  an optional file, terminate the fallback chain with `|| true` and gate on\n  `[[ -n $VAR ]]` (see shell-scripting skill).\n\n## Related\n\n- `watchers` skill — feed polling with watermark dedup (RSS\u002FJSON\u002FGitHub\n  issues); use for \"new items\" streams. This skill is for \"did ONE specific\n  bug get fixed\" — marker dedup instead of watermark, silent-until-fixed\n  semantics instead of new-item emission.\n- `fleet-commands` — UPDATE_REQUEST dispatch + authoritative-bus verification.\n","# Upstream Fix Watchdog\n\nWhen an upstream project (e.g. hermes-agent by NousResearch) ships a bug that\nforces you to **pin a version and pause fleet update crons**, you need a cheap\nwatchdog that tells you the moment the fix lands — without nagging every day\nwhile it's still broken.\n\nPattern proven 2026-08-25→28: `_stdio_children_dead()` inversion in\nhermes-agent's `tools\u002Fmcp_tool.py` broke every stdio MCP call on all hosts.\nFleet pinned, paused `agent-hermes-update`, and a daily cron\n(`check-hermes-upstream-fix`) watched upstream until the fix appeared, then\nalerted with the exact fleet-action steps. Generalize this for every future\nupstream pin.\n\n## When to use\n\n- Upstream bug forces a pin\u002Fpause; you want to know when it's safe to unpin\n- A dependency you don't control may silently keep a known-bad state\n- You need a \"wake me up the day this changes\" watch on a raw upstream file\n- User asks for \"a watcher for an upstream fix\" \u002F \"notify me when X is fixed\"\n\n## Core design (all non-negotiable)\n\n1. **Silent while bug present** — exit 0 with EMPTY stdout. `no_agent=true`\n   cron watchdog pattern: empty stdout = silent tick, no delivery, no tokens.\n2. **Notify exactly once** — marker file dedupes:\n   `MARKER = ~\u002F.hermes\u002Fstate\u002F\u003Cincident>-notified`; write it after first\n   notify; skip future notifies if it exists.\n3. **Fetch failure = LOUD error** — network\u002FHTTP errors print\n   `CHECK FAILED: ...` to stderr and exit 1, so a broken check can never go\n   silent. A silent watchdog that stopped polling is worse than no watchdog.\n4. **Conservative detection** — when the upstream file's structure changes so\n   the marker can't be parsed, treat as \"not fixed\" (return False), never as\n   \"fixed\". False-negative → one more day of silence; false-positive → wakes\n   the fleet for nothing.\n5. **Concrete buggy-marker anchor** — match the exact inverted line (e.g.\n   `return True  # alive`), PLUS positive proof of the fix (e.g. the\n   corrected `return False` path present, or the function deleted entirely).\n\n## Script skeleton\n\n```python\n#!\u002Fusr\u002Fbin\u002Fenv python3\n\"\"\"Watchdog: notify when upstream \u003Crepo> fixes \u003Cbug>. Silent until then.\"\"\"\nimport os, re, sys, urllib.request\n\nDEFAULT_URL = \"https:\u002F\u002Fraw.githubusercontent.com\u002F\u003Corg>\u002F\u003Crepo>\u002Fmain\u002F\u003Cpath>\"\nMARKER = os.path.expanduser(\"~\u002F.hermes\u002Fstate\u002F\u003Cincident>-notified\")\nBUGGY_MARKER = \"\u003Cexact buggy line>\"\nNOTIFY_TEXT = \"\"\"🟢 UPSTREAM FIX DETECTED: \u003Cbug> is fixed on origin\u002Fmain.\nFleet action:\n  1. \u003Cunpin\u002Fupdate steps>\n  2. \u003Cresume paused crons: hermes cron resume \u003Cjob_id>>\n  3. \u003Cverify: grep -c '\u003Cbuggy-symbol>' \u003Cfile> on each host>\n  4. \u003Cremove the pin once all hosts verify clean>\"\"\"\n\ndef fetch(url: str) -> str:\n    req = urllib.request.Request(url, headers={\"User-Agent\": \"steadfaste-fleet-fix-watchdog\"})\n    with urllib.request.urlopen(req, timeout=30) as resp:\n        return resp.read().decode(\"utf-8\", errors=\"replace\")\n\ndef is_fixed(src: str) -> bool:\n    # 1. Function\u002Fsymbol absent entirely -> bug gone\n    # 2. Extract the function body; if BUGGY_MARKER in body -> not fixed\n    # 3. Else require positive proof (correct return path present)\n    raise NotImplementedError  # per-incident logic\n\ndef main(argv):\n    url = argv[0] if argv and argv[0].startswith(\"http\") else DEFAULT_URL\n    dry = \"--dry-run\" in argv\n    try:\n        src = fetch(url)\n    except Exception as exc:\n        print(f\"CHECK FAILED: could not fetch {url}: {exc}\", file=sys.stderr)\n        return 1\n    if not is_fixed(src):\n        return 0  # silent\n    if not dry and os.path.exists(MARKER):\n        return 0  # already notified\n    print(NOTIFY_TEXT)\n    if not dry:\n        os.makedirs(os.path.dirname(MARKER), exist_ok=True)\n        open(MARKER, \"w\").write(\"notified\\n\")\n    return 0\n\nif __name__ == \"__main__\":\n    sys.exit(main(sys.argv[1:]))\n```\n\nFull working example: `~\u002F.hermes\u002Fscripts\u002Fcheck-hermes-upstream-fix.py`\n(2026-08-25 incident — real `is_fixed()` regex, marker handling, notify text).\n\n## Cron wiring\n\n```\nhermes cron create --name \u003Cname> --schedule '0 9 * * *' \\\n  --script \u003Cname>.py --no-agent --deliver origin\n```\n\n- `no_agent=true` — script stdout IS the delivery; empty = silent tick\n- Daily 09:00 is a sane cadence for upstream fixes (they don't land hourly)\n- Test before enabling: `python3 \u003Cscript>.py --dry-run` (must print the\n  notify text but NOT write the marker), then `python3 \u003Cscript>.py` once with\n  the marker deleted to see a real tick\n- When the fix fires and the fleet converges, **pause the cron**\n  (`cronjob(action='pause')`) — the job is done; don't let it tick\n  pointlessly forever. Keep the script for the next incident.\n\n## When the fix fires — fleet convergence playbook\n\n1. **Verify on the authoritative source first** — `git fetch origin && git\n   log --oneline -1 origin\u002Fmain` or curl the raw file and eyeball the diff.\n   The watchdog is a tripwire, not proof.\n2. **Unpin \u002F update on the orchestrator host first** (esther\u002Fmoses), verify\n   the service actually works (not just \"command exited 0\").\n3. **Dispatch the fleet — ALWAYS, without asking** (Luke directive\n   2026-08-28: *\"always send for these types of updates. you never have to\n   ask and it just wastes time actually\"*). Do NOT pause to ask permission\n   for a fix\u002Fupdate dispatch — it is the default, not a decision point.\n   `orch-bus-fleet-dispatch.py --execute` (sync stale local\n   agent-registry from the bus host first — see `fleet-commands` skill), or\n   `fleet send \u003Cagent> UPDATE_REQUEST ...`.\n   Include ALL agents; the dispatch tool skips dev-agents — send those\n   manually. When the registry's `health_url` is a stale placeholder\n   (e.g. `your-domain.com`), `fleet send` liveness refuses — use `--force`\n   (the bus is the authoritative proof; verify against `bus.archives`).\n   Verify UPDATE_RESULT `git_sha_after` on the AUTHORITATIVE bus\n   (never the local mirror — it lags).\n4. **Resume paused crons** — `hermes cron resume \u003Cjob_id>` for each\n   `agent-hermes-update`-class job paused during the incident.\n5. **Verify per host** — `grep -c '\u003Cbuggy-symbol>' \u003Cfile>` = 0, or confirm\n   the fixed logic.\n6. **Pause the watchdog** (see Cron wiring above).\n\n## Upstream update may ask \"restore local changes now?\" — decide by content\n\nDuring `hermes update` the updater autostashes local changes and asks whether\nto restore. Decide by content, not by default:\n\n- `git stash show --stat stash@{0}` — see what's inside (working tree is\n  clean when the stash exists, so status won't help)\n- **Local workarounds are DELIBERATE here** (fleet preference): lean skill\n  index, cost guard, max-cost preflight. Restore them — they're your\n  environment protection, not accidental edits\n- Check upstream first: `git show HEAD:\u003Cfile> | grep -c \u003Csymbol>` — if\n  upstream now HAS the functionality (e.g. a `lean` mode), the local patch\n  may be partially redundant; restore anyway (semantics may differ)\n- Conflict check: `git rev-list --count \u003Cstash-base>..HEAD -- \u003Cfiles>` — low\n  count = safe; `git stash show -p stash@{0}` inspects WITHOUT mutating\n- ⚠️ **`git stash apply` MUTATES the working tree** — it is NOT a dry run.\n  To inspect: `git stash show -p stash@{0}`. Only `apply`\u002F`pop` when you\n  intend to restore. (Learned 2026-08-28: an \"inspection\" apply created a\n  real conflict mid-update; restored via\n  `git restore --source=HEAD --staged --worktree \u003Cfiles>` + `git rm --cached`\n  for untracked stash files.)\n\n## Pitfalls\n\n- **The upstream \"fix\" may be incomplete — grep sibling call sites (2026-08-28).**\n  The `_stdio_children_dead()` inversion was fixed upstream, but the same\n  feature's `_watch_stdio_children` probe still had its own coroutine-leak\n  bug: `inspect.isawaitable(_watch_children())` INVOKES the async function,\n  creating a never-awaited coroutine per MCP call (`RuntimeWarning: coroutine\n  'MCPServerTask._watch_stdio_children' was never awaited`). On Titus this\n  made loop-governance MCP calls fail silently — `begin_change` never\n  completed, so the enforcer blocked ALL write tools \"even for inspection.\"\n  **Before trusting an upstream fix, grep the whole feature for the same\n  flaw class** — a fix to one function often leaves siblings broken. See\n  `references\u002Fmcp-tool-watch-probe-case.md` for the full reproduction.\n- **Seam-rot on the LOCAL patch re-apply (whole-deploy abort).** Steadfaste Fleet\n  re-applies local hermes-agent patches after every `hermes update` via\n  `ops\u002Fscripts\u002Finstall\u002Finstall-*.py` (lean index, cron cost-tracking, mcp-tool\n  watch). When upstream refactors the seam, the installer's marker anchor no\n  longer matches → it prints `FAIL ... anchor not found` and exits 2, which\n  aborts `fleet-update.sh` under `set -euo pipefail` BEFORE the state-save\n  writes `state\u002Fupdate-commit`. Net: the deploy looks normal but the doctor's\n  `Deploy sync` stays red forever. Diagnose by mechanism (`fleet-update.sh`\n  exit 0 is NOT clean — grep the log for `[\u003Cinstaller>] FAIL ... anchor not\n  found`, compare `state\u002Fupdate-commit` to `git rev-parse HEAD`). Distinguish\n  load-bearing (re-target the anchor — e.g. keeping `coding_context: lean`\n  distinct when upstream folds it into `focus`) from superseded (mechanism\n  removed upstream — convert to treat \"absent upstream\" as success, not\n  drift-fail), then re-target in REPO source, verify markers match, COMMIT (a\n  concurrent workflow can `git reset` and sweep uncommitted edits), then\n  deploy. Verify behavior with a live import (`_coding_mode({'coding_context':\n  'lean'})` must return `lean`), not just exit 0.\n- **Marker file must be per-incident** — reuse the same `\u003Cname>` for a new\n  incident and the old marker suppresses the new notify. Name it after the\n  incident (`upstream-hermes-fix-notified`), not generically.\n- **Raw URL vs git clone** — `raw.githubusercontent.com` is the cheap fetch\n  (single file, no auth, no clone). Only clone when you need the whole tree.\n- **Don't trust `return True`\u002F`False` alone** — the exact inverted line is\n  the anchor; the fix's positive return path is the proof. Stubs that remove\n  all returns are NOT fixes.\n- **Delivery target `origin`** — one alert to the operator's DM is the\n  point; don't fan out to every agent.\n- **Keep the script after the incident** — the next upstream break is a\n  5-minute edit (URL + marker + notify text), not a new build.\n- **`set -euo pipefail` assignment trap** — if the script is bash and reads\n  an optional file, terminate the fallback chain with `|| true` and gate on\n  `[[ -n $VAR ]]` (see shell-scripting skill).\n\n## Related\n\n- `watchers` skill — feed polling with watermark dedup (RSS\u002FJSON\u002FGitHub\n  issues); use for \"new items\" streams. This skill is for \"did ONE specific\n  bug get fixed\" — marker dedup instead of watermark, silent-until-fixed\n  semantics instead of new-item emission.\n- `fleet-commands` — UPDATE_REQUEST dispatch + authoritative-bus verification.\n",{"name":5,"tags":24,"version":32,"description":7,"related_skills":33},[25,26,27,28,29,30,31],"cron","watchdog","upstream","fix","fleet","pin","monitoring","1.0.0",[34,35,36,37],"watchers","fleet-commands","cron-format-standard","steadfaste-fleet-maintenance","main","2e6aa8d15ab0a366c90be54f61bf9c6e97015e83","skills\u002Fdevops\u002Fupstream-fix-watchdog",1,"https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet","steadfaste.com","https:\u002F\u002Fgithub.com\u002Flukemcqueen","AI Architect | Building Hermes Cortex: governed multi-agent AI for autonomous software engineering | Enterprise Tech Consultant","2026-11-14 22:33:34",{"source":8,"skillName":5,"commands":48,"prompt":58,"zipAvailable":59},[49,52,55],{"tool":50,"command":51},"npx","npx skills add lukemcqueen\u002Fsteadfaste-fleet --skill upstream-fix-watchdog",{"tool":53,"command":54},"bunx","bunx skills add lukemcqueen\u002Fsteadfaste-fleet --skill upstream-fix-watchdog",{"tool":56,"command":57},"pnpm","pnpm dlx skills add lukemcqueen\u002Fsteadfaste-fleet --skill upstream-fix-watchdog","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fupstream-fix-watchdog\u002FSKILL.md",true,[],[62,70,77,84,91,97],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":69},"01M7DXY4G8NSVB59K8B6S2AQNP","macbook-linux-maintenance","Legacy MacBook Linux fixes: GPU wedge, reboot hang, thermal.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fmacbook-linux-maintenance\u002FSKILL.md",[],[],"2026-11-15 14:19:17",{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":69},"01M7DXXYJCXFJ3TGB84ZQ8KFE6","governance-regression-triage","Triage a host's governance issues: what changed.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fgovernance-regression-triage\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":69},"01M7DXXSAP1G8FDRJAMXQGBBMK","enforcement-immutability","Use when enforcement files lose their immutable flag.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fenforcement-immutability\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":69},"01M78SSDQVRYX64Y16X32TC0NS","steadfaste-fleet-repo-deploy","Use when committing to the steadfaste-fleet repo.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fsteadfaste-fleet-repo-deploy\u002FSKILL.md",[],[],{"skillNo":92,"name":37,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":69},"01M78SS928F3R2RJAYVW4E30CE","Maintain an installed Steadfaste Fleet instance — update both the upstream Hermes Agent and the fleet repo layer, merge diverged histories, sync skills safely, and troubleshoot common issues. Covers the dual-update workflow (hermes update + fleet-update.sh), encrypted brain data backup, mycortex import, and recovery techniques.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fsteadfaste-fleet-maintenance\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":69},"01M78SP1CHW0T0BWE5ZT6JBPX8","fleet-preflight","DEPRECATED alias — merged into survey-before-action (2026-08-20). Load survey-before-action instead; its \"Repo-Specific Pre-Flight\" section covers git search, Hermes boundary, and deployment verification.","https:\u002F\u002Fgithub.com\u002Flukemcqueen\u002Fsteadfaste-fleet\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Ffleet-preflight\u002FSKILL.md",[],[]]