| name | research-refresh |
| description | Audits freshness across all briefs. Lists briefs by age class — current (<30d), aging (30-90d), stale (>90d). Optionally re-runs stale briefs via /deep-research. Use for weekly or monthly maintenance, or before a big /create session where multiple topics will be referenced. Auto-triggers on: "/research-refresh", "check brief freshness", "which briefs are stale", "audit the research cache", "refresh stale briefs". Do not use for a specific topic — use /deep-research <topic> for that.
|
| allowed-tools | Read, Glob, Bash(date *) |
Research Refresh
Audit and optionally refresh stale briefs.
Inputs
- Optional:
--auto-refresh — re-run briefs whose age exceeds the threshold. Default: report only.
- Optional:
--threshold <days> — override the 30-day default for the stale classifier.
Execution
Step 1: Scan
Glob state/briefs/*/claims.json. For each, read the freshness field. Compute age = today − freshness.
Step 2: Classify
| Age | Class | Default action |
|---|
| < 30 days | current | skip |
| 30 - 90 days | aging | flag |
| > 90 days | stale | flag; re-run only if --auto-refresh |
Also classify anything with verdict: FAIL as failed — surface these regardless of age.
Step 3: Report
<N> briefs total:
current: <count>
aging: <count>
- <slug> (<age> days, verdict: <X>)
stale: <count>
- <slug> (<age> days, verdict: <X>)
failed: <count>
- <slug> (verdict: FAIL — re-run needed)
Step 4: Conditional refresh
If --auto-refresh was passed:
- For each
stale brief, invoke /deep-research <slug> sequentially (not parallel — avoid fetch storms).
- Report per-brief outcome:
<slug>: <old verdict> → <new verdict>.
failed briefs are NOT auto-refreshed — they may have an underlying issue that re-running won't fix. Ask the developer.
Rules
- Do NOT auto-refresh unless
--auto-refresh is explicitly passed.
- Do NOT delete stale briefs. Retention is the developer's call.
- Do NOT re-run
aging briefs automatically — 30-90 days may be an intentional snapshot window.
- Run refreshes sequentially, not in parallel — the fetcher's WebSearch budget is shared.
Failure modes
| Code | Signature | Counter |
|---|
| F10 | Auto-deleted a brief as "cleanup" | This skill never deletes; retention is manual |
| F09 | Parallel re-runs caused fetch races | Serialize auto-refreshes |
