Add monthly artifact-audit skill + second maintenance schedule - #11
Merged
Conversation
Second half of the maintenance answer. update-knowledge (weekly) keeps the knowledge base current with the world; this adds artifact-audit (monthly), the inward-facing complement that checks the repo's OWN skills/templates/ guardrails/runbooks for drift against that knowledge base and proposes fixes as a human-reviewed PR. Key design points: - Distinct task, not a chunk of the weekly: outward research vs. inward consistency, different cadence (artifacts drift slower), different output (edits practice artifacts, never knowledge/). - No auto-apply / no self-edit privilege — that fast-path stays reserved to update-knowledge's own CI-fenced file. Everything artifact-audit proposes is an ordinary human-reviewed PR edit, and it prefers flagging over rewriting anything touching a hard stop, the mission, permissions, or the self-edit gate. - Same one-shot guardrail envelope (bounded fan-out, subagent cap, no recursion, human merge, claude/-prefixed branch). runbooks/staying-current.md gains a "two maintenance loops (and why not five)" section documenting the cadence analysis: keep each job single (chunking the weekly research by angle would fragment one coherent artifact into conflicting PRs); the two schedules are separate because they're different tasks, not because work was split for size. README updated. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Nsk6ryTqYCAuE6dRW8ZnHJ
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this is (PR 2 of 2)
The maintenance-structure half of your second question. PR #10 broadened the repo beyond Claude Code; this PR makes sure the repo's own artifacts stay correct as the tools drift — and answers the "should the weekly be chunked?" question.
Independent of PR #10 — they share no files, so they merge in any order.
The cadence answer (in the runbook)
Don't chunk the weekly research by angle. Its five angles produce one coherent artifact (deduped and cross-referenced against each other, one backlog carry-forward, one CHANGELOG entry, one review). Splitting into five schedules → five PRs all editing
primer.md/sources.md/CHANGELOG.mdin the same week → merge conflicts between your own PRs (the exact failure the step-1 open-PR check was added to prevent), 5× review load, lost synthesis. Keep each job single; if a weekly delta ever feels too big, go twice-weekly — not angle-split.But there was a real gap: "maintaining the techniques" is broader than the knowledge base. Nothing scheduled kept the repo's own skills / templates / guardrails / runbooks correct as tools drift. That's a distinct task, so it gets a distinct (slower) schedule — not more knowledge-research schedules.
What's added
New skill
artifact-audit— the inward-facing complement toupdate-knowledge:update-knowledge(weekly)artifact-audit(monthly)knowledge/knowledge/)It reads the KB as source of truth, fans out ~4 bounded agents to audit skills/templates/guardrails/runbooks for drift, verifies each finding against a High-confidence KB entry or primary source, and opens a PR with Fixed + Flagged for you sections.
Safety:
artifact-audithas no auto-apply / noself-edit:privilege — that fast-path stays reserved toupdate-knowledge's own CI-fenced file. Everything here is an ordinary human-reviewed PR edit, and it prefers flagging over rewriting anything touching a hard stop, the mission, permissions,budget.env, or the self-edit gate. It never editsknowledge/. Same one-shot guardrail envelope (bounded fan-out, subagent cap, no recursion, human merge,claude/-prefixed branch).runbooks/staying-current.md— a "two maintenance loops (and why not five)" section documenting the cadence analysis, plus the monthly Routine setup (points at the skill file fresh, same anti-staleness pattern as the weekly). README updated.Coordination between the two routines
Each checks for an already-open PR of its own kind (anti-conflict), and
artifact-auditexplicitly yields to an openupdate-knowledgePR on any file it's mid-editing (notablyupdate-knowledge/SKILL.md).Needs you (settings-only)
Set up the second Routine (monthly) per the runbook — same as the weekly with a monthly trigger. The env caps / overage-off guidance applies to both.
Not merging — human reviews and merges.
Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01Nsk6ryTqYCAuE6dRW8ZnHJ
Generated by Claude Code