| name | nack |
| description | Re-check a prior assistant claim. Use when user says "nack", "not convinced", or "check again"; not for review feedback. |
NACK
Negative acknowledgement: treat the user's pushback as a request to re-check, not as hostility or approval.
Quick start
- Identify the exact prior claim, recommendation, or action being challenged.
- Re-check the strongest evidence available: prior context, docs, live files, tests, or commands.
- Report whether the answer changes, stands, or is blocked by missing evidence.
- If a safe correction is obvious, make the smallest correction and verify it.
Operational basis
Inspect only what bears on the challenged point:
- the latest user pushback and the assistant message it targets;
- repo instructions and
git status --short --branch before edits;
- named files/docs/tests/commands behind the original claim;
codebase-map-understand.md only for cross-module or architecture claims, then verify named files in live source.
Evaluate external reviewer feedback against live code and tests before applying it. Use grill-me or grill-with-docs for full plan stress tests.
Workflow
- State
Rechecking: <specific claim>.
- Look for disconfirming evidence first.
- Decide:
Changed: original claim was wrong or too strong; correct it.
Stands: original claim still holds; cite the evidence briefly.
Blocked: the evidence needed is unavailable; ask one focused question or name the command/approval needed.
- Continue only with the corrected safe next action.
Skill contract
Entry protocol
- Clear target: re-check directly.
- Ambiguous target: re-check the latest substantive assistant claim.
- Multiple risky interpretations: ask which claim/action to challenge.
Topology check
Before editing, confirm ownership, blast radius, validation signal, and whether the correction changes prior user intent.
Verification gate
A useful NACK pass includes the challenged claim, evidence checked, verdict, and any validation for changes made.
Red lines
Do not treat pushback as approval to broaden scope, rewrite history, deploy, spend money, expose secrets, or make destructive changes. Do not argue from confidence; use evidence.
Output contract
Rechecking: <claim/action>
Verdict: Changed|Stands|Blocked — <one-line reason>
Evidence: <files/commands/context checked>
Next: <small safe action or question>
Example
User: nack, not convinced. check again
Agent: re-checks the latest recommendation against the repo, then either corrects it or says why it still stands with evidence.
Shared contract
Follow the shared skill contract for repo study, dirty-worktree hygiene, verification evidence, safe handoffs, and safety defaults.