fix

Apply code changes and fixes. Writes implementation code, applies bug fixes, and verifies changes with tests. Core action hub in the development mesh.

SKILL.md

GitHub

fix

Purpose

Apply code changes. Fix receives a plan, debug finding, or review finding and writes the actual code. It does NOT investigate root causes — that is rune:debug's job. Fix is the action hub: locate, change, verify, report.

<HARD-GATE> Never change test files to make tests pass unless the tests themselves are provably wrong (wrong expected value, wrong test setup, testing a removed API). The rule: fix the CODE, not the TESTS. If unsure whether the test is wrong or the implementation is wrong → call `rune:debug` to investigate. </HARD-GATE>

Triggers

  • Called by cook Phase 4 IMPLEMENT — write code to pass tests
  • Called by debug when root cause found and fix is ready
  • Called by review when bugs found during review
  • /rune fix <issue> — manual fix application
  • Auto-trigger: after successful debug diagnosis

Calls (outbound)

  • debug (L2): when root cause unclear before fixing — need diagnosis first
  • test (L2): verify fix with tests after applying changes
  • review (L2): self-review for complex or risky fixes
  • verification (L3): validate fix doesn't break existing functionality
  • docs-seeker (L3): check correct API usage before applying changes
  • hallucination-guard (L3): verify imports after code changes
  • scout (L2): find related code before applying changes
  • neural-memory (L3): after fix verified — capture fix pattern (cause → solution)
  • adversary (L2): on agent.stuck after 2+ failed attempts — oracle-mode dispatches stateless second-model pass to break confirmation-bias loop

Called By (inbound)

  • cook (L1): Phase 4 IMPLEMENT — apply code changes
  • debug (L2): root cause found, ready to apply fix
  • review (L2): bug found during review, needs fixing
  • surgeon (L2): apply refactoring changes
  • review-intake (L2): apply fixes identified during structured review intake
  • graft (L2): apply integration fixes for grafted code
  • scaffold (L1): apply fixes during project scaffolding

Cross-Hub Connections

  • fix ↔ debug — bidirectional: debug diagnoses → fix applies, fix can't determine cause → debug investigates
  • fix → test — after applying fix, run tests to verify
  • fix ← review — review finds bug → fix applies correction
  • fix → review — complex fix requests self-review

Execution

Step 1: Understand

Read and fully understand the fix request before touching any file.

  • Read the incoming request: debug report, plan spec, or review finding
  • Identify what is broken or missing and what the expected behavior should be
  • If the request is ambiguous or root cause is unclear → call rune:debug before proceeding
  • Note the scope: single function, single file, or multi-file change

Step 1b: Recovery Policy Matrix

Before locating code, classify the incoming error/task into a recovery category to determine the right fix strategy. This prevents wasting effort on the wrong approach.

Error Type Recovery Action Strategy
INPUT_REQUIRED — missing user input, ambiguous spec PROMPT_USER Return NEEDS_CONTEXT with specific questions. Do NOT guess.
INPUT_INVALID — wrong format, type mismatch, encoding AUTO_FIX Fix at validation layer. Add schema validation (Zod/Pydantic) if missing.
TIMEOUT — operation exceeded time limit RETRY with adjustment Increase timeout, add retry with exponential backoff, or chunk the operation.
POLICY_BLOCKED — security gate, lint rule, contract violation ABORT Do NOT work around the policy. Report to caller with the specific rule that blocked.
PERMISSION_DENIED — auth failure, file access, API scope PROMPT_USER Cannot fix permissions programmatically. Report exact permission needed.
DEPENDENCY_ERROR — missing package, version conflict, broken dep AUTO_FIX Install missing dep, resolve version conflict, or suggest alternative package.
LOGIC_ERROR — wrong output, incorrect calculation, bad algorithm INVESTIGATE Do NOT auto-fix. Call rune:debug — logic errors need root cause analysis.
ENVIRONMENT_ERROR — wrong Node/Python version, missing system dep PROMPT_USER Report exact version/tool needed. Agent cannot change system environment.

Decision flow:

  1. Read the incoming diagnosis/error
  2. Classify into one of the 8 error types above
  3. Apply the recovery action — this determines whether to proceed (AUTO_FIX, RETRY), ask (PROMPT_USER), stop (ABORT), or re-diagnose (INVESTIGATE)
  4. Announce: "Recovery policy: {error_type} → {action}"

Why: Without a recovery matrix, fix attempts the same strategy (read → change → test) for every error type. A POLICY_BLOCKED error doesn't need code reading — it needs the policy reported. An INPUT_REQUIRED error doesn't need debugging — it needs a question asked. Matching strategy to error type eliminates wasted cycles.

Step 2: Locate

Find the exact files and lines to change.

  • Use rune:scout to locate the relevant files, functions, and surrounding code
  • Use Read to examine the specific file:line identified in the debug report or plan
  • Use Glob to find related files: types, tests, config that may also need updating
  • Map all touch points before writing a single line of code

Step 3: Change

Apply the minimal set of changes needed.

  • Use Edit for targeted modifications to existing files
  • Use Write only when creating a genuinely new file is required
  • Follow project conventions: naming, immutability patterns, error handling style
  • Keep changes minimal — fix the stated problem, do not refactor unrelated code (YAGNI)
  • Never use any in TypeScript; never use bare except: in Python
  • If a new import is needed → note it for Step 5 hallucination-guard check

Step 4: Verify

Confirm the change works and nothing is broken.

  • Use Bash to run the relevant tests: the specific failing test first, then the full suite
  • If tests fail after the fix:
    • Investigate with rune:debug (max 3 debug loops before escalating)
    • Do NOT change test files to make tests pass — fix the implementation code
  • If project has a type-check command, run it via Bash
  • If project has a lint command, run it via Bash

Step 4.5: Quality Decay Check (Self-Regulation)

When fix is called repeatedly (e.g., by cook Phase 4, or iterative fix loops), track a WTF-likelihood score — the probability that continued fixing is making things worse.

Compute every 3 fix attempts (or when called 5+ times in a single cook session):

Signal Score Adjustment
A fix was reverted (any test that passed now fails) +15%
Fix touched >3 files (blast radius expanding) +5% per extra file beyond 3
15+ fixes already applied in this session +1% per fix beyond 15
All remaining issues are LOW severity +10%
Fix touched files outside the original diagnosis scope +20%
Consecutive fixes without running tests between them +10%

Thresholds:

  • >20% WTF-likelihood: STOP fixing. Report current state to cook/user with: "Quality decay detected — continued fixes risk introducing more bugs than they resolve. {N} fixes applied, {score}% risk. Recommend: commit current progress, re-assess remaining issues."
  • Hard cap: 30 fixes per session — regardless of score. After 30, STOP and report.
  • 2+ consecutive fixes on the same file all failed: emit agent.stuck signal. scout zoom-out mode (structural pivot) and adversary oracle-mode (semantic pivot via stateless second-model dispatch) both listen and fire in parallel. If oracle.response arrives with confidence=high, apply its recommended edit (still routes through normal validation gates).

Reset conditions: WTF-likelihood resets to 0% when:

  • User explicitly says "continue fixing"
  • A full test suite run shows zero regressions
  • Scope is narrowed to a single file

Step 5: Post-Fix Hardening (Defense-in-Depth)

After the fix works, make the bug structurally impossible — not just "fixed this time."

Single validation at one point can be bypassed by different code paths, refactoring, or mocks. Add validation at EVERY layer data passes through:

Layer Purpose Example
Entry Point Reject invalid input at API boundary Validate params not empty/exists/correct type
Business Logic Ensure data makes sense for this operation Check preconditions specific to this function
Environment Guard Prevent dangerous ops in specific contexts In tests: refuse writes outside tmpdir
Debug Instrumentation Capture context for forensics if bug recurs Log stack trace + key values before risky ops

Apply this when: the bug was caused by invalid data flowing through multiple layers. Skip for trivial one-liner fixes.

Step 5b: Preserve Debug Instrumentation

If rune:debug left #region agent-debug markers in the code:

  1. During fix: DO NOT remove these markers — they capture the investigation trail
  2. After fix verified (tests pass, lint pass): scan for #region agent-debug markers
  3. Remove markers and their contents in a final cleanup pass ONLY after full verification
  4. If the fix is partial or tests still fail → KEEP all markers for the next debug cycle

Why: Premature cleanup of debug instrumentation erases failure history. If the bug recurs after cleanup, the next debug session starts from zero. Keeping markers until verification means downstream skills can see what was already investigated.

Step 6: Self-Review

Verify correctness of the changes just made.

  • Call rune:hallucination-guard to verify all imports introduced or modified are real and correctly named
  • Call rune:docs-seeker if any external API, library method, or SDK call was added or changed
  • For complex or risky fixes (auth, data mutation, async logic): call rune:review for a full quality check

Step 6b: Capture Fix Pattern

Call neural-memory (Capture Mode) to save the fix pattern: what broke, why, and how it was fixed. Priority 7 for recurring bugs.

Step 7: Report

Produce a structured summary of all changes made.

  • List every file modified and a one-line description of what changed
  • Include verification results (tests, types, lint)
  • Note any follow-up work if the fix is partial or has known limitations

Constraints

  1. MUST NOT change test files to make tests pass — fix the CODE, not the TESTS
  2. MUST have a diagnosis (from debug or clear error) before applying fixes
  3. MUST run tests after each fix attempt — never batch multiple untested changes
  4. MUST NOT exceed 3 fix attempts — if 3 fixes fail, re-diagnose via rune:debug (which will classify: wrong approach → brainstorm rescue, wrong design → plan redesign)
  5. MUST follow project conventions found by scout — don't invent new patterns
  6. MUST NOT add unplanned features while fixing — fix only what was diagnosed
  7. MUST track fix attempt number — this feeds debug's 3-Fix Escalation classification
  8. MUST preserve #region agent-debug markers until fix is fully verified — cleanup only after tests pass

Scope Gate

Change Type Action
Bug fix (diagnosed cause) Fix it
Security fix (found during fix) Fix it + flag to sentinel
Blocking issue (can't complete fix without) Fix it + document in report
Unrelated improvement STOP — create separate task
Architectural change STOP — escalate to cook/plan

If fix requires touching >3 files not in the diagnosis → re-diagnose. You're probably fixing a symptom.

Mesh Gates

Gate Requires If Missing
Evidence Gate Debug report OR clear error description before fixing Run rune:debug first
Test Gate Tests run after each fix attempt Run tests before claiming fix works

Output Format

## Fix Report
- **Task**: [what was fixed/implemented]
- **Status**: DONE | DONE_WITH_CONCERNS | NEEDS_CONTEXT | BLOCKED

### Changes
- `path/to/file.ts` — [description of change]
- `path/to/other.ts` — [description of change]

### Verification
- Lint: PASS | FAIL
- Types: PASS | FAIL
- Tests: PASS | FAIL ([n] passed, [m] failed)

### Concerns (if DONE_WITH_CONCERNS)
- [concern]: [impact assessment] — [suggested remediation]

### Context Needed (if NEEDS_CONTEXT)
- [what is unknown]: [why it blocks] — [two most likely answers]

### Blocker (if BLOCKED)
- [specific blocker]: [what was attempted]

### Notes
- [any caveats or follow-up needed]

Status Protocol (Subagent Contract)

Fix returns one of four statuses to its caller (cook, debug, review, surgeon). The caller uses this to route next actions.

Status When Example
DONE Fix applied, tests pass, no issues Clean bug fix, all green
DONE_WITH_CONCERNS Fix works but has side effects or caveats worth noting "Tests pass but performance regressed 15% — consider optimizing in follow-up"
NEEDS_CONTEXT Cannot apply fix without clarification — ambiguous spec or missing info "Two valid interpretations of the expected behavior — need user input"
BLOCKED Hard blocker — exhausted fix attempts, broken dependency, fundamental incompatibility "3 fix attempts failed — triggering debug escalation"

Returns

Artifact Format Location
Code changes Source files Per debug report / plan file paths
Fix Report Markdown (inline) Emitted to calling skill (cook, debug, review, surgeon)
Verification output Inline (Fix Report) Lint + types + test results

Chain Metadata

Append to Fix Report when invoked standalone. Suppress when called as sub-skill inside an L1 orchestrator (cook, team, etc.) — the orchestrator emits a consolidated block. See docs/references/chain-metadata.md.

chain_metadata:
  skill: "rune:fix"
  version: "1.0.0"
  status: "[DONE | DONE_WITH_CONCERNS | NEEDS_CONTEXT | BLOCKED]"
  domain: "[area fixed]"
  files_changed:
    - "[list of modified files]"
  exports:
    fix_applied: { files: ["[paths]"], description: "[what was fixed]" }
    verification: { lint: "[PASS/FAIL]", types: "[PASS/FAIL]", tests: "[PASS/FAIL]" }
    commit_hash: "[hash if committed]"
  suggested_next:
    - skill: "rune:test"
      reason: "[grounded in changes — e.g., 'Modified 3 files in auth module, edge cases need coverage']"
      consumes: ["fix_applied", "verification"]

Sharp Edges

Known failure modes for this skill. Check these before declaring done.

Failure Mode Severity Mitigation
Modifying test files to make tests pass CRITICAL HARD-GATE blocks this — fix the code, never the tests (unless test setup is provably wrong)
Applying fix without a diagnosis HIGH Evidence Gate: need debug report or clear error description before touching code
Exceeding 3 fix attempts without re-diagnosing HIGH Constraint 4: after 3 failures, call debug again — the hypothesis was wrong
Introducing unrelated refactoring while fixing MEDIUM YAGNI: fix only what was diagnosed — unrelated changes belong in a separate task
Not running tests after each individual change MEDIUM Constraint 3: never batch untested changes — run tests after each edit
Fixing at crash site without tracing data origin HIGH Defense-in-depth: trace where bad data ORIGINATES, add validation at every layer it passes through
Single-point validation (fix one spot, hope it holds) MEDIUM Step 5: add entry + business logic + environment + debug layers for data-flow bugs
Removing debug instrumentation before fix is verified MEDIUM Step 5b: preserve #region agent-debug markers until all tests pass — premature cleanup erases failure history
Runaway fix loop — 20+ fixes without checking quality decay HIGH Step 4.5: WTF-likelihood self-regulation. >20% risk = STOP. Hard cap 30 fixes/session. Each fix adds risk — diminishing returns after ~15
Each fix creates a new bug elsewhere — whack-a-mole CRITICAL Tight coupling signal. STOP fixing → escalate to debug with note "each fix creates new failure — suspect structural issue". Debug will route to plan for redesign
Applying same fix strategy to every error type MEDIUM Step 1b Recovery Policy Matrix: classify error type FIRST — POLICY_BLOCKED needs reporting not fixing, INPUT_REQUIRED needs questions not code

Done When

  • Root cause identified (debug report or clear error received)
  • Minimal changes applied targeting only the diagnosed problem
  • Tests pass for the fixed functionality (actual output shown)
  • Lint and type check pass
  • hallucination-guard verified any new imports
  • Fix Report emitted with 4-state status, changed files, and verification results
  • If DONE_WITH_CONCERNS: concerns listed with impact + remediation
  • If NEEDS_CONTEXT: specific questions stated with two likely answers
  • If BLOCKED: blocker + all attempted approaches documented

Cost Profile

~2000-5000 tokens input, ~1000-3000 tokens output. Sonnet for code writing quality. Most active skill during implementation.

Scope guardrail: Do not refactor unrelated code or create new features beyond the diagnosed fix target unless explicitly delegated by the parent agent.