| name | fix |
| description | Fix issues from a failed Strike verification pass, then return to the same verifier. |
| argument-hint | [failed verification context] |
| disable-model-invocation | true |
| allowed-tools | Read Write Edit MultiEdit Bash Grep Glob WebFetch WebSearch Agent |
Fix
Fix issues from a failed Strike verification pass.
Inputs
- failed verification artifact
- current initiative, phase, or slice artifacts named by the verifier
- prior fix reports for the same verifier, when present
- Required user-provided customization from the consuming repo's Strike
workspace:
- user implementation-discipline guidance:
strike/user-guidance/implementation-discipline/global.md
strike/user-guidance/implementation-discipline/fix.md
- repo files, commands, or paths needed to understand and repair the issue
Process
- Read the failed verification artifact first.
- Read required user-provided customization and apply relevant guidance to the
repair:
strike/user-guidance/implementation-discipline/global.md
strike/user-guidance/implementation-discipline/fix.md
- Fix the verifier's
Must Fix items. They may be broken behavior,
incomplete accepted scope, missing evidence, weak artifacts, or regressions.
- Treat
P0 and P1 Must Fix items as required before the verifier can
pass. Treat P2 and P3 findings as follow-up unless the verifier says they
are required by accepted scope.
- Read whatever project docs, artifacts, code, or source-backed references are
needed to fix the real issue.
- Before patching a technical symptom, search the codebase for how this repo
already handles the same class of problem. Use error terms, provider names,
object names, workflow step names, storage/upload paths, and nearby adapters or
callers as search anchors.
- If the repair creates or modifies utilities, helpers, adapters, schemas, or
shared modules, search for existing repo patterns and inspect likely callers
before editing.
- For integration, provider, workflow, upload, asset, storage, queue, job,
callback, webhook, or dataflow repairs, name the existing repo precedent used
or record that none was found before making code changes.
- Keep the repair focused on passing the failed verification, not expanding the
feature.
- Treat
Follow-Up and Accepted Risk items as context. Do not fix them unless
they are necessary to resolve a Must Fix item.
- Review prior fix reports for the same verifier. If the same issue repeats
after a reasonable repair attempt, write
Fixed: no with the repeated issue
instead of broadening the work.
- Run focused checks that prove the repair, plus any repo-level check that is
appropriate for the changed surface.
- When the repair requires a real user, product, data, security, accounting,
scope, or architecture decision, write
Fixed: no, explain the needed
decision, and return to the verifier or route-back flow without inventing the
answer.
- When the repair would edit audited phase
research.md or
research-audit.md, write Fixed: no with Route Back to
phaseResearchComplete so Strike reruns the phase research step and its audit.
- If the issue belongs to an earlier workflow step instead of a contained
repair, write
Fixed: no and fill in ## Route Back.
Output
Write a compact fix report next to the affected workflow evidence.
Use the next unused numeric path:
strike/initiatives/<initiative-id>/phases/<phase-id>/slices/<slice-id>/fixes/fix-001.md
strike/initiatives/<initiative-id>/phases/<phase-id>/fixes/fix-001.md
strike/initiatives/<initiative-id>/fixes/fix-001.md
Use the most local path that matches the failed verification.
Use this shape:
# Fix
## Source Verification
Verifier:
Artifact:
## Fix Status
Fixed: yes / no
Reason:
## Issues Addressed
- I1 —
## Changes Made
-
## Repo Precedent
Searched:
Closest existing pattern:
How the repair follows or intentionally differs:
## Checks
-
## Not Fixed / Blockers
- None.
## Follow-Up / Accepted Risk
- None.
## Return
Verifier:
Artifact:
Ready for re-verification: yes / no
## Route Back
Needed: yes / no
Command: None / reopen-check / reopen-phase-check / reopen-slice-check
Phase: None / <phase-id>
Slice: None / <slice-id>
Check: None / <state-check>
Reason:
Rules
- Fix only from a failed verification artifact.
- Use one of three outcomes:
Fixed: yes and return to the same verifier;
Fixed: no with Route Back when an earlier workflow step owns the repair;
or Fixed: no without route-back only for a genuinely unresolved decision.
- When writing route-back, use the exact helper command Strike should run.
Use
reopen-check for the current scope, reopen-phase-check for a phase
check, and reopen-slice-check for a specific slice check.
- Do not decide that the work is verified. The same verifier must run again.
- Do not complete workflow state checks.
- Do not chase open-ended improvements.
- Do not treat a failed check, provider response, workflow error, payload limit,
upload/storage issue, or dataflow mismatch as novel until you have searched
the repo for existing handling of that class of problem.
- Do not hide accepted-scope defects as follow-up work.
- Do not rewrite the failed verification artifact; it remains the source for
this repair loop.
- Preserve unrelated user work.