| name | weekly-ops-audit |
| description | Weekly automated audit of the ops architect agent harness — runs every Sunday evening alongside weekly-pipeline-review |
You are auditing [YOUR_NAME]'s Claude Code agent harness for the past week. Pre-parsed system data is provided below as JSON — do NOT read files yourself.
Input Data
The following JSON was pre-computed by fetch-audit-data.py:
{{AUDIT_DATA_JSON}}
Step 0 — 4Cs System Health (run before harness checks)
Quick read-only check of system context — takes 2 minutes, surfaces drift before it causes problems.
Context — Is the CLAUDE.md and context still accurate?
- Does
context/current-priorities.md reflect this week's actual focus?
- Does
context/work.md reflect current employment status?
- Is
context/goals.md still pointing at the right targets?
- Flag anything that reads stale or contradicts recent sessions.
Connections — Are integrations healthy?
- Read
context/connections.md — are there any ⚠️ or ❌ status entries?
- Were any MCP tools failing or timing out this week (check error events in audit data)?
- Is Infisical injection still working for scheduled tasks?
- Update
connections.md Status column for anything that changed this week.
Capabilities — Are skills earning their place?
- (Covered in Step 1 skill coverage section below)
- Also flag: any skills in registry.json that haven't been used in 3+ weeks
Cadence — Is the rhythm still working?
- Are scheduled tasks running on their intended schedule (check logs)?
- Any tasks that fired but produced no useful output?
- Is the weekly audit itself running consistently?
Add a 4Cs Summary section to the Notion report:
Context: OK / [what needs updating]
Connections: OK / [what's broken or stale]
Capabilities: [N unused skills flagged]
Cadence: OK / [what's off-schedule]
Step 1 — Run these checks using the JSON above
Hook health
- Are all 3 hooks (pre-tool.sh, post-tool.sh, stop.sh) present and executable?
- Did any
budget_halt events fire this week? If yes, how many and in which sessions?
- Did any
budget_warning events fire? Flag if warnings are frequent (>3/session on average).
Permission drift
- Are there tools appearing in
tool_considered events that are NOT listed in registry.json? If yes, list them — they may need to be added to the registry.
- Are there any
permission_denied events? List the tool and session.
Skill coverage
- Which skills from registry.json were used this week (appear in
tool_selected events)?
- Which skills were NOT used at all? Flag any that have been unused for 2+ consecutive weeks (note: you only have 1 week of data per run — flag if unused this week and add a note to monitor).
Budget health
- Average tool calls per session this week.
- Sessions that hit the warning threshold (60 calls).
- Sessions that hit the hard limit (80 calls).
Error signals
- Any
error events in the log? List them with their session_id and timestamp.
Step 3 — Write the gap report
Create a Notion page under "My Job Search" parent (page ID: 2ae58ea23b8e81b0a4bbf9a99ea6a538) titled "Weekly Ops Audit — [date]" with the following sections:
System Health
- Hooks: OK / Issues found
- Permissions: OK / Drift detected
- Budget: Average X tool calls/session, N warnings, N halts
What Needs Attention
(Only include if issues found — skip section if all clear)
- List each issue with: what it is, why it matters, what to do
Skill Usage This Week
- Used: [list]
- Unused: [list] — monitor next week
Primitive Coverage (rate each as: Healthy / Needs work / Not yet implemented)
- Tool registry
- Permission tiers
- Session logging
- Budget tracking
- Weekly audit (this one)
- Deferred: workflow state machine, idempotency, streaming events, multi-agent coordination, memory system
One Recommendation
The single highest-leverage improvement to make before next Sunday's audit.
Keep everything scannable — bullets, not paragraphs.
After creating the page, notify with:
osascript -e 'display notification "Weekly ops audit is ready in Notion" with title "Ops Audit" sound name "Glass"'
