| name | repair |
| description | Internal dynos-work skill. Manually repair a specific finding. Use when you want to fix one issue without running the full audit loop. /dynos-work:audit handles repair automatically. |
dynos-work: Repair
Manually repair a specific finding from an audit report. Use this when you need to fix a precise issue without running the full audit pipeline.
Ruthlessness Standard
- A repair that cannot survive re-audit is not a repair.
- Fix the mechanism, not the screenshot of the symptom.
- If the instruction is vague, stop and sharpen it before spawning an executor.
What you do
- Find the active task in
.dynos/
- Read the latest audit reports
- If the user specifies a finding ID, repair that finding only
- If no finding specified, show all open blocking findings and ask which to repair
- Spawn the appropriate executor subagent with the precise repair instruction
- After repair, re-run only the auditor(s) that reported the finding (plus always spec-completion and security)
- Report new audit result
Usage
/dynos-work:repair — shows all open findings, ask which to fix
/dynos-work:repair sec-003 — repairs finding sec-003
/dynos-work:repair --all — repairs all open findings in parallel (where file-safe)
Executor selection
Use the finding's assigned_executor from the audit report. If it is missing, stop and surface the gap instead of inferring ownership from file extension or guessing.
Hard rules
- Always re-audit after repair — do not assume the fix worked
- Always include spec-completion and security in the re-audit
- Do not hand-write
repair-log.json; that file belongs to the deterministic audit repair loop
