Agent 7: Scheduled Agent
| Property | Value |
|---|---|
| Tier | Utility (Scan) / Production (Analysis) |
| Trigger | Weekly cron schedule |
| Purpose | Proactive health monitoring of the library |
| Est. Cost | 5k - 15k tokens per weekly scan |
Weekly Scans
1. Dependency Staleness
- uses: actions/setup-python@v5
- run: |
pip install pip-audit safety
pip-audit --format=json > audit.json
safety check --json > safety.json
Claude Sonnet creates issues for:
- Dependencies with known CVEs (priority: high)
- Dependencies > 2 major versions behind (priority: medium)
- Dependencies with abandoned upstream (priority: low, flag for replacement)
2. Compatibility Matrix Check
- Are we testing against the latest Python release?
- Has a new version been released that we should add to the matrix?
- Are any versions in our matrix approaching EOL?
3. Tech Debt Scan
# TODO/FIXME/HACK comment count
grep -rn "TODO\|FIXME\|HACK" src/ --include="*.py" | wc -l
# Cyclomatic complexity
radon cc src/ -n C -j # functions with complexity C or worse
4. Community Health Report
- GitHub stars trend (up/down/flat)
- Issue velocity (opened vs closed per week)
- PR merge time (getting slower?)
- Download stats from PyPI
- Unanswered issues older than 14 days
5. Confidence-Gated Workload Entry
The Scheduled Agent does not auto-create tasks. Instead:
Scan finds issue
→ Agent assigns confidence score (0-100%)
→ Confidence > 80%: Draft issue with label "scheduled:auto"
→ Confidence 50-80%: Draft issue with label "scheduled:review-needed"
→ Confidence < 50%: Added to weekly summary for human, no issue created
The human reviews the weekly summary and promotes items to the workload queue.