audit 2026-10-23: scheduled-tasks window (172/0, 5 incidents closed); reopen skills-index drift; skills-index generator now excludes .archive (306 skills, was 347)

This commit is contained in:
Tony0410
2026-10-23 00:14:49 +08:00
parent 1d92de31a0
commit 59d4b2e695
6 changed files with 712 additions and 419 deletions
+54 -4
View File
@@ -1,14 +1,15 @@
---
title: "Skills Index Drift — 4x Undercount"
type: incident
status: resolved
status: active
created: 2026-08-15
updated: 2026-10-21
updated: 2026-10-23
resolved_on: 2026-08-15
verified_on: 2026-10-21
reopened_on: 2026-10-23
verified_on: 2026-10-23
confidence: high
tags: [skills, wiki, drift, auto-regeneration]
sources: [session-investigation, agent-estate-wiki]
sources: [session-investigation, agent-estate-wiki, live-filesystem-probe, cron-executions-db]
---
# Skills Index Drift — 4x Undercount
@@ -48,6 +49,55 @@ nested deeper). The daily cron job `1a8e90ea2271` is `active` with `last_status:
- Page includes "auto-regenerated, do not hand-edit" warning
- Same script pattern is reusable for other inventory pages (cron jobs, services, agents)
## Reopened 2026-10-23 — the count includes archived skills, and two generators disagree
The 2026-10-21 all-clear above was premature. Two defects were confirmed on the
night of 2026-10-22/23:
**1. The stocktake ran six times long and committed nothing.** Cron job
`1a8e90ea2271` ran **03:00:28 → 04:38:19 AWST on 2026-10-22 (97 m 58 s)**, against
**16 m 01 s** on 2026-10-21, and finished `completed` with delivery `suppressed` —
but `git status` in the wiki repo still shows the run's regenerated
`systems/skills-index.md` (mtime 10-22 03:00) **uncommitted**, and none of the fixes
the run had diagnosed were applied. A `completed` row is therefore not evidence that
the stocktake did its job.
**2. Two scripts generate the same page with different shapes.**
| Generator | Skill scan | 2026-10-22 output |
|---|---|---|
| `/home/hermes/workspace/agent-estate-wiki/wiki-maintain.sh` (the one the cron job executes) | `find … -name SKILL.md` — does **not** prune dot-directories | **347** skills, with a `## .Archive (41)` section |
| `/home/hermes/.hermes/scripts/wiki-maintain.sh` (stale copy, mtime 2026-08-17) | `Path.rglob('SKILL.md')` | **348** skills, no archive section |
The 00:18 maintenance step committed the `rglob` shape; the 03:00 stocktake wrote the
`find` shape. Whoever runs last wins, so the page oscillates.
**Live counts, verified 2026-10-23 00:1x AWST** (filesystem probes + `hermes skills list`):
- `find ~/.hermes/skills -name SKILL.md` → **348**
- `find -L ~/.hermes/skills -name SKILL.md` → **376** (28 more skills reachable only through symlinks)
- excluding `.archive/` (**41**) and `apple/` (**4**) → **331**
- the loader's own catalogue — `hermes skills list` → **329 enabled** (`34 hub-installed, 52 builtin, 243 local — 329 enabled, 0 disabled`)
So the page total counts **45 artefacts the loader does not expose** (the parked
`.archive/` set plus `apple/`), which is why its number has never matched the loader.
**Fix applied 2026-10-23:** the repo's `wiki-maintain.sh` now excludes `*/.archive/*`
from the scan, prints the archived count as an explicit exclusion, and states that
`hermes skills list` is the authoritative *loadable* figure while the page remains a
filesystem inventory. The page was regenerated through the patched script and
committed with the fix.
**Still open:**
- The stale `~/.hermes/scripts/wiki-maintain.sh` copy must never be run — it produces
the divergent shape. It should be deleted or made to delegate to the repo script.
- The page will still read higher than the loader's 329: nested collections
(`context-engineering-collection/…`), `apple/` and hub-installed trees are counted as
installed files. Whether the page should instead report the loader's figure is a
decision for Anthony.
- Watch the **2026-10-23 03:00** stocktake run: it should complete in ~16 minutes and
commit the regenerated page.
## Related
- [[systems/skills-index]] — The updated page
- [[systems/scheduled-tasks]] — The new cron job