refactor— agent skill

<objective>

Test-first refactoring. Audit coverage, add characterization tests if missing, apply changes with safety net.

NOT for:

• bug fixes (use /develop:fix)
• new features (use /develop:feature)
• .claude/ config changes (use /foundry:manage (requires foundry plugin))
• non-Python projects (JS/TS/Go/Rust) — toolchain assumes pytest; use language-native toolchain instead
• mixed refactor+feature tasks — run /develop:refactor first, then /develop:feature; do not attempt both in single skill run

Quality stack (Branch Safety Guard, Codex Pre-pass, Progressive Review) requires foundry plugin; when absent, Step 5 quality stack is skipped with a visible warning — output is lower quality but workflow still completes.

</objective>

<constants>

• MAX_INNER_CYCLES: 5 (change-test cycles per outer session — Step 4 safety break)

</constants>

<workflow>

<!-- Agent Resolution: resolved at runtime via $_DEV_SHARED; source at develop/skills/_shared/agent-resolution.md (installed path) -->

Agent Resolution

_PATHS=$(python "${CLAUDE_PLUGIN_ROOT:-plugins/develop}/bin/dev_shared_resolve.py" --foundry 2>/dev/null)  # timeout: 5000
_DEV_SHARED=$(echo "$_PATHS" | head -1)
_FOUNDRY_SHARED=$(echo "$_PATHS" | tail -1)
[ -z "$_FOUNDRY_SHARED" ] && _FOUNDRY_SHARED="plugins/foundry/skills/_shared"

Read $_DEV_SHARED/agent-resolution.md. Contains: foundry check + fallback table. If foundry not installed: use table to substitute each foundry:X with general-purpose. Agents skill uses: foundry:sw-engineer, foundry:qa-specialist, foundry:linting-expert, foundry:challenger.

Read $_DEV_SHARED/task-hygiene.md.

<!-- Reference only — execution-dead at runtime; included for agent behavioral context -->

Anti-Rationalizations

Project Detection

Read $_DEV_SHARED/runner-detection.md — sets $TEST_CMD (full suite) and $PYTEST_CMD (pytest flags). Run at skill start.

Optional `--plan <path>`: if $ARGUMENTS contains --plan <path> (at any position), read plan file first. Extract Affected files, Risks, Suggested approach — use to inform Step 1 scope analysis. Skip redundant codebase exploration for already-classified files. Store plan path as PLAN_FILE.

Read $_DEV_SHARED/preflight-helpers.md — execute --plan path extraction; sets $PLAN_FILE.

Checkpoint init: run DEV_DIR=$(python "${CLAUDE_PLUGIN_ROOT:-plugins/develop}/bin/dev_run_dir.py" 2>/dev/null) # timeout: 5000 to create .developments/<TS>/ and capture path. Write checkpoint.md inside $DEV_DIR. After each major step (1, 2, 3, 4, 5), append step: N — completed to $DEV_DIR/checkpoint.md. On skill start, check for existing .developments/*/checkpoint.md — offer resume from last completed step if found.

Flag parsing

Parse flags into actual shell variables (not prose) so downstream blocks see correct values. Persist to temp files for cross-block access (bash state lost between Bash() calls):

# timeout: 10000
python "${CLAUDE_PLUGIN_ROOT:-plugins/develop}/bin/dev_parse_args.py" \
    --skill refactor --write-files "$ARGUMENTS"

Downstream blocks read back, e.g. TEAM_MODE=$(cat ${TMPDIR:-/tmp}/dev-team-mode 2>/dev/null || echo false).

Codemap flag parsing — derive raw flag into a real shell variable, then normalize via codemap-resolve. Uses skill-specific temp file (dev-refactor-codemap-raw) to avoid reading stale values from prior feature/debug runs:

# timeout: 5000
CODEMAP_RAW=auto
[[ " $ARGUMENTS " == *" --no-codemap "* ]] && CODEMAP_RAW=off
[[ " $ARGUMENTS " == *" --codemap "* ]] && [[ " $ARGUMENTS " != *" --no-codemap "* ]] && CODEMAP_RAW=strict
echo "$CODEMAP_RAW" > ${TMPDIR:-/tmp}/dev-refactor-codemap-raw

Unsupported flag check — after all supported flags extracted, scan $ARGUMENTS for remaining --<token> tokens. If found: print ! Unknown flag(s): \--<token>\. Supported: \--plan\, \--team\, \--no-challenge\, \--codemap\, \--no-codemap\, \--accept-no-plan\, \--semble\, \--repo\. then invoke AskUserQuestion — (a) Abort (stop, re-invoke with correct flags) · (b) Continue ignoring (skip unknown flags, proceed). On Abort: stop.

Codemap auto-detection — run after flag parsing. Behaviour differs by mode: strict (user explicitly passed --codemap) hard-fails when codemap unavailable; auto and off soft-degrade to false (do not abort skill):

# timeout: 5000
CODEMAP_RAW=$(cat ${TMPDIR:-/tmp}/dev-refactor-codemap-raw 2>/dev/null || echo auto)
CODEMAP_ENABLED=$("${CLAUDE_PLUGIN_ROOT:-plugins/develop}/bin/codemap-resolve" "$CODEMAP_RAW")
RESOLVE_EXIT=$?
if [ "$RESOLVE_EXIT" -ne 0 ]; then
    if [ "$CODEMAP_RAW" = "strict" ]; then
        echo "! codemap unavailable but --codemap (strict) passed — aborting"
        exit 1
    fi
    # auto/off: soft degrade — skill continues without codemap
    echo "⚠ codemap unavailable in '$CODEMAP_RAW' mode — proceeding with CODEMAP_ENABLED=false"
    CODEMAP_ENABLED=false
fi
echo "$CODEMAP_ENABLED" > ${TMPDIR:-/tmp}/dev-refactor-codemap-enabled
# codemap: integrated-via-shared

Preflight — if CODEMAP_ENABLED=true:

Read $_DEV_SHARED/preflight-helpers.md — execute codemap + semble preflight if respective flags set.

Step 1: Scope and understand

Read target code, build mental model before touching anything.

If <target> is directory: use Glob tool (pattern **/*.py, path <target>) to enumerate Python files.

find <target> -name '*.py' -exec wc -l {} + 2>/dev/null | tail -1

If `CODEMAP_ENABLED=true` or `SEMBLE_ENABLED=true`: read $_DEV_SHARED/codemap-context.md and follow enabled sections (codemap block if CODEMAP_ENABLED, semble companion if SEMBLE_ENABLED). Skip if both false.

Multi-file / API-change scope — extended codemap scan (only when CODEMAP_ENABLED=true): if target is directory, spans multiple files, or goal mentions renaming/restructuring public API (i.e., refactoring NOT limited to internals of single function or class with unchanged public interface):

PROJ=$(basename "$(git rev-parse --show-toplevel 2>/dev/null)")  # timeout: 3000
REFACTOR_FILES=$(find <target> -name '*.py' -type f 2>/dev/null)
AFFECTED_MODULES=$(echo "$REFACTOR_FILES" | sed 's|^\./||;s|^src/||;s|\.py$||;s|/|.|g' | grep . || echo "")
_IDX="${CODEMAP_INDEX_DIR:-.cache/codemap}"
if command -v scan-query >/dev/null 2>&1 && [ -f "${_IDX}/${PROJ}.json" ] && [ -n "$AFFECTED_MODULES" ]; then
    while IFS= read -r mod; do
        scan-query rdeps "$mod" 2>/dev/null
    done <<< "$AFFECTED_MODULES"
    scan-query coupled --top 10
fi

Include ## Scope & Reusability (codemap) block in foundry:sw-engineer spawn prompt. If rdeps returns callers outside refactoring scope: flag explicitly — those callers must update or refactoring silently breaks public contract. If CODEMAP_ENABLED=false and scope is multi-file: skip silently.

Spawn foundry:sw-engineer agent to analyze code and identify:

• Public API surface (functions, classes, methods external code calls)
• Internal complexity hotspots (cyclomatic complexity, deep nesting, long functions)
• Code smells relevant to stated goal
• Dependencies and coupling between modules
• Complexity smell: directory or cross-module scope — flag it; consider team mode

Read $_DEV_SHARED/premise-grounding.md §Premise Grounding Gate. Apply using refactor context from the Skill contexts table.

Goal classification gate: after sw-engineer analysis completes, scan goal text for mixed signals — if goal contains both refactor keywords (rename, extract, restructure, decouple, consolidate) AND feature keywords (add, implement, new, support), invoke AskUserQuestion: "Goal mixes refactoring and feature work — split into two runs." · (a) Abort — run refactor first, then feature · (b) Continue as refactor-only — treat feature additions as out of scope.

Scope gate: if target spans 3+ modules OR 5+ files OR goal mentions any public-API rename — flag complexity smell. Use AskUserQuestion: "Narrow scope (Recommended)" / "Proceed anyway".

Read $_DEV_SHARED/plan-inline.md §Inline Plan Generation Protocol. Apply using refactor context from the Skill contexts table. On proceed: set PLAN_FILE=<path>; continue to Step 2. On small complexity or ACCEPT_NO_PLAN=true: skip and continue to Step 2.

Challenger gate

Skip if `CHALLENGE_ENABLED=false`.

Spawn foundry:challenger with scope analysis from Step 1 (affected files, dependencies, coupling, risks):

"Review the refactoring scope and approach. Challenge across all 5 dimensions: Assumptions, Missing Cases, Security Risks, Architectural Concerns, Complexity Creep. Apply mandatory refutation step."

Parse result:

• Blockers found → STOP. Present findings. Don't proceed to Step 2 until user resolves each blocker or explicitly accepts risk.
• Concerns only → surface as advisory before coverage audit; continue.
• No findings / all refuted → proceed.

Step 2: Audit test coverage

Find existing tests for target code:

Use Glob tool (pattern **/test_*.py or **/*_test.py), then Grep tool (pattern <module_name>, output mode files_with_matches) to narrow to those referencing target.

(Use Glob tool — pattern: **/test_*.py — to discover test files; check pyproject.toml [tool.pytest.ini_options] testpaths for configured paths)

# timeout: 600000
$PYTEST_CMD --co -q 2>&1 | head -5

SKIP_COV=0
if $PYTEST_CMD --co -q --cov=. 2>&1 | grep -q "ModuleNotFoundError\|No module named.*cov"; then
    echo "⚠ coverage tool not found — coverage gate skipped"
    SKIP_COV=1
fi

$PYTEST_CMD --co -q 2>&1 | grep -i "<module_name>" || echo "No tests found for <module_name>"

[ "${SKIP_COV}" -eq 0 ] && { $PYTEST_CMD --cov=<target_module> -q --cov-report=term-missing || true; }

If SKIP_COV=1: skip coverage classification entirely — do not classify any function as UNCOVERED; note "coverage tool absent — coverage audit skipped" in the audit output. Step 3 qa-specialist spawn behavior when `SKIP_COV=1`: spawn qa-specialist with all public functions listed as coverage: unknown and instruction to write characterization tests for every public function (cannot prioritize uncovered functions when coverage unknown — test all to ensure safety net). Proceed to Step 3 with unknown coverage state.

Classify each public function/method (only when SKIP_COV=0):

• Covered: at least one test for happy path + one edge case
• Partially covered: test exists but missing edge cases or failure paths
• Uncovered: no test

Review: Validate the coverage audit

Before writing characterization tests, evaluate audit output critically:

1. Completeness: all public functions, methods, classes identified — including complex call paths?
2. Classification accuracy: each item correctly classified? Partial-covered often misclassified as covered.
3. Refactor relevance: uncovered/partial items in code paths refactoring will touch?
4. Hidden dependencies: integration points or cross-module calls audit may have missed?

If audit incomplete: re-examine before Step 3. Gaps found mid-refactoring (Step 4) costly.

<!-- Only active when --team flag passed (~10% of invocations) --> Team mode branch — if TEAM_MODE=true: Steps 1–2 complete solo (teammates need scope + coverage context). Spawn both teammates now; skip Steps 3–5, proceed to Final Report after results received.

When TEAM_MODE=true:

Compute run directory and create health sentinel:

# timeout: 5000
_run=$(python "${CLAUDE_PLUGIN_ROOT:-plugins/develop}/bin/setup_worktree.py" --sentinel refactor-team-check)
TS=$(echo "$_run" | head -1)
RUN_DIR=$(echo "$_run" | tail -1)
RUN_DIR_LITERAL="$RUN_DIR"
echo "$TS" > ${TMPDIR:-/tmp}/dev-refactor-team-ts
echo "$RUN_DIR" > ${TMPDIR:-/tmp}/dev-refactor-run-dir
trap 'rm -f ${TMPDIR:-/tmp}/refactor-team-check-$TS' EXIT

IMPORTANT: in spawn prompts below, substitute $RUN_DIR_LITERAL with the actual resolved path before constructing each Agent call — agents receive literal resolved strings, not shell variable references. Same applies to $TS substitution.

Note on `model=` assignments: model=opus/model=sonnet in the spawn prompts below are advisory hints — effective only when the actual foundry agents are installed. When falling back to general-purpose (foundry absent), the prompt-prepend model= does not reliably override the agent-resolution fallback tier; effective model is set by agent-resolution.md's fallback table, not the spawn prompt. This is intentional — sonnet is sufficient for the qa-specialist characterization-test task and opus for sw-engineer refactor implementation; on fallback, expect tier degradation noted in Final Report.

Serialize teammates — sw-engineer applies refactor first, then qa-specialist writes characterization tests against the post-refactor source. Running in parallel produces a race: qa-specialist would write tests against the pre-refactor implementation and they would fail once sw-engineer completes. qa-specialist spawn prompt must include: "Read $RUN_DIR_LITERAL/refactor-sw-engineer.md for the refactored source before writing any tests — do NOT test pre-refactor source files."

Step T1 — Spawn foundry:sw-engineer (model=opus) and wait for completion. Prompt: "You are a foundry:sw-engineer teammate refactoring: [target]. Read ~/.claude/TEAM_PROTOCOL.md — use AgentSpeak v2. Your task: apply the refactoring steps (Steps 4–5: change with safety net, review). Scope constraint: only edit source files (not under tests/). Broadcast context: {target: <path>, coverage: <summary>, goal: <stated goal>}. Compact Instructions: preserve file paths, test results, coverage numbers. Discard verbose tool output. Task tracking: do NOT call TaskCreate or TaskUpdate — the lead owns all task state. Signal completion in final delta message: 'Status: complete | blocked — <reason>'. Write your full analysis to $RUN_DIR_LITERAL/refactor-sw-engineer.md using the Write tool. Return ONLY compact JSON: {\"status\":\"done\",\"file\":\"<path>\",\"findings\":N,\"confidence\":0.N,\"summary\":\"<one-line>\"}."

Step T2 — Only after T1 returns success, spawn foundry:qa-specialist (model=sonnet) against the refactored source. Prompt: "You are a foundry:qa-specialist teammate refactoring: [target]. Read ~/.claude/TEAM_PROTOCOL.md — use AgentSpeak v2. Your task: write characterization tests (Step 3) to build a safety net for the refactor — test the CURRENT (post-refactor) source produced by Teammate 1. Scope constraint: only create/edit files under tests/. Do NOT edit source files. Broadcast context: {target: <path>, coverage: <summary>, goal: <stated goal>, refactor_output: $RUN_DIR_LITERAL/refactor-sw-engineer.md}. Compact Instructions: preserve file paths, test results, coverage numbers. Discard verbose tool output. Task tracking: do NOT call TaskCreate or TaskUpdate — the lead owns all task state. Signal completion in final delta message: 'Status: complete | blocked — <reason>'. Write your full analysis to $RUN_DIR_LITERAL/refactor-qa-specialist.md using the Write tool. Return ONLY compact JSON: {\"status\":\"done\",\"file\":\"<path>\",\"findings\":N,\"confidence\":0.N,\"summary\":\"<one-line>\"}."

Health monitoring (CLAUDE.md §6): re-derive $TS and $RUN_DIR at block start (bash state lost between Bash() calls — read back from temp files the spawn block persisted):

# timeout: 5000
TS=$(cat ${TMPDIR:-/tmp}/dev-refactor-team-ts 2>/dev/null || date -u +%Y-%m-%dT%H-%M-%SZ)
RUN_DIR=$(cat ${TMPDIR:-/tmp}/dev-refactor-run-dir 2>/dev/null || echo ".temp/develop/$TS")

Apply to each teammate independently — create sentinel touch ${TMPDIR:-/tmp}/refactor-team-check-$TS before each spawn; every 5 min: find $RUN_DIR -newer ${TMPDIR:-/tmp}/refactor-team-check-$TS -type f | wc -l — new files = alive; zero = stalled. Hard cutoff: 15 min no file activity → timed out. One extension (+5 min) if tail -20 of output file explains delay; second unexplained stall = hard cutoff. On timeout: read tail -100 of stalled file; surface partial results with ⏱; never omit.

After both complete: read their output files from $RUN_DIR/, synthesize outputs, run quality stack, produce Final Report. Exit — do not continue to Steps 3–5.

Continue to Step 3 only when TEAM_MODE=false.

Step 3: Add characterization tests (if needed)

For every uncovered or partially covered public API, spawn foundry:qa-specialist to generate characterization tests:

• Import function, call with representative inputs, assert current output
• Use pytest.mark.parametrize for multiple input/output pairs
• Name tests test_<function>_characterization_*

Spawn with context:

• Target module: <module_path>
• Coverage audit results: [paste coverage-audit output showing uncovered/partial functions]
• Uncovered public APIs to test: [list from audit]
• Current code (read target file before writing tests — tests must assert CURRENT behaviour, not desired)
• Test file target: tests/test_<module>_characterization.py
• Test naming: test_<function>_characterization_<scenario>

# timeout: 600000
$PYTEST_CMD <test_file> -v; GATE_EXIT=$?
echo "$GATE_EXIT" > ${TMPDIR:-/tmp}/dev-gate-exit

Gate: all characterization tests must pass before proceeding. Check exit code from persisted file ($? in a fresh shell is unrelated to the prior pytest run):

GATE_EXIT=$(cat ${TMPDIR:-/tmp}/dev-gate-exit 2>/dev/null || echo 1)
if [ "${GATE_EXIT}" -eq 5 ]; then
    echo "GATE FAIL: no tests collected (exit 5) — characterization test file missing or not detected by pytest; cannot proceed to Step 4 without a safety net"
elif [ "$GATE_EXIT" -ne 0 ]; then
    echo "GATE FAIL: characterization test(s) failed (exit $GATE_EXIT) — fix the test, not the code"
else
    echo "GATE OK: all characterization tests pass on unmodified code"
fi

If GATE_EXIT -ne 0 (including exit 5): characterization tests are missing or wrong — cannot proceed to Step 4 without a passing safety net. Invoke AskUserQuestion — "Characterization test gate failed (exit $GATE_EXIT). How to proceed?" · (a) Fix test collection path / fix test assertions (recommended — re-spawn qa-specialist with corrected path or assertions) · (b) Proceed without safety net (accept risk — record decision in $DEV_DIR/checkpoint.md) · (c) Abort. On (b): document explicit acceptance in checkpoint.md (step: 3 — gate exit $GATE_EXIT — proceed without safety net (user accepted)) before continuing.

Step 4: Refactor with safety net

For each change:

1. One focused change (single responsibility per edit)
2. Run affected tests (prefer targeted over full characterization suite):

   scan-query test-impact "<changed_module>" 2>/dev/null

• Non-empty pytest_cmd → run those tests; surface not_covered caveat if present; fall back to full suite if all tests pass but feel incomplete
• Empty or unavailable → full suite:

   # timeout: 600000
   $PYTEST_CMD --tb=short <test_files> -v

1. Tests pass: proceed to next change
2. Tests fail: revert, try different approach

Safety break: track cycle count and wall time via temp files (bash state lost between Bash() calls — $INNER_CYCLE and $START_TIME declared inline are unavailable in subsequent Bash blocks; persistence is mandatory):

# timeout: 3000
echo "0"             > ${TMPDIR:-/tmp}/dev-inner-cycle
echo "$(date +%s)"   > ${TMPDIR:-/tmp}/dev-start-time
MAX_WALL_SECONDS=1800  # 30 min hard cap (5 outer × MAX_INNER_CYCLES inner worst case)

At each inner iteration start, read back, increment, check:

# timeout: 3000
INNER_CYCLE=$(cat ${TMPDIR:-/tmp}/dev-inner-cycle 2>/dev/null || echo 0)
START_TIME=$(cat ${TMPDIR:-/tmp}/dev-start-time 2>/dev/null || echo $(date +%s))
INNER_CYCLE=$((INNER_CYCLE+1))
echo "$INNER_CYCLE" > ${TMPDIR:-/tmp}/dev-inner-cycle
MAX_INNER_CYCLES=5  # must match constants block — bash cannot reference it directly
if [ "$INNER_CYCLE" -gt $MAX_INNER_CYCLES ]; then
    echo "⚠ MAX_INNER_CYCLES ($MAX_INNER_CYCLES) reached — stopping refactor loop; report what succeeded, what broke, what remains"
fi
ELAPSED=$(( $(date +%s) - START_TIME ))
if [ "$ELAPSED" -ge 1800 ]; then
    echo "⚠ wall-time cap reached (30 min) — stopping refactor loop"
fi

After each change-test pair: re-read counter from temp file, increment, write back. Stop when INNER_CYCLE > MAX_INNER_CYCLES or elapsed ≥ MAX_WALL_SECONDS.

Refactoring categories:

• Logic simplification: replace complex conditionals, flatten nesting, extract helpers
• API cleanup: rename for clarity, consolidate parameters, add type annotations
• Structural: extract classes/modules, reduce coupling, apply design patterns
• Performance: replace loops with vectorized ops, reduce allocations, batch I/O
• Dead code removal: remove unused imports, unreachable branches, commented-out code; scan _-prefixed functions with no call sites; flag public methods absent from __init__.py exports

Step 5: Review and close gaps

Full review of refactored code. Loop — review -> targeted refactoring (return to Step 4) -> re-review until only nits remain. Max 3 outer cycles. (Step 4's "max 5 change-test cycles" bound applies within each pass through Step 4, independent of outer loop.)

Each cycle:

1. Evaluate against all criteria:

• Behavior preservation: all characterization tests and pre-existing tests pass with identical outputs
• Goal achieved: stated refactoring goal actually accomplished (not just partial)
• No new smells: no new coupling, complexity, or duplication introduced
• API surface: no unintended public API changes (signature, return type, raised exceptions)
• Dead code: unreachable code after refactor was removed

1. For every gap: return to Step 4, apply targeted fix — one focused change per gap.

1. Re-run full test suite:

   # timeout: 600000
   $PYTEST_CMD --tb=short <test_files> -v 2>&1 | tail -20
   GATE_EXIT=${PIPESTATUS[0]}

1. Objective convergence check: if findings this cycle identical to previous cycle (same locations, same issues), declare convergence and exit — further cycles won't resolve; surface to user.

1. Only nits remain (variable naming, comment clarity, minor formatting): document in Follow-up, exit loop.

1. Substantive gaps remain: start next cycle (max 3 total).

After 3 cycles: substantive issues remain → stop, surface to user.

Foundry availability check before quality stack:

# timeout: 3000
[ ! -d "$_FOUNDRY_SHARED" ] || [ ! -f "$_FOUNDRY_SHARED/quality-stack.md" ] && echo "⚠ foundry plugin not installed — quality stack skipped (Branch Safety Guard, Codex Pre-pass, Progressive Review Loop, Codex Mechanical Delegation)"

Read $_FOUNDRY_SHARED/quality-stack.md (if not found → skip quality stack entirely, note "foundry plugin absent — quality stack skipped (Branch Safety Guard, Codex Pre-pass, Progressive Review, Codex Mechanical Delegation)" in Final Report) and execute Branch Safety Guard, Quality Stack, Codex Pre-pass, Progressive Review Loop, and Codex Mechanical Delegation steps.

Final Report

## Refactor Report: <target>

### Goal
[stated goal or "general quality pass"]

### Test Coverage Before
- Covered: N functions | Partially: N | Uncovered: N
- Characterization tests added: N

### Changes Made
| File | Change | Lines |
|------|--------|-------|
| path/to/file.py | extracted helper function | -12/+8 |

### Test Results
- All tests passing: yes/no
- Coverage: before% -> after%

### Follow-up
- [any remaining items that need manual review]

## Confidence
**Score**: 0.N — [high ≥0.9 | moderate 0.85–0.9 | low <0.85 ⚠]
**Gaps**:
- [e.g., coverage tool unavailable, some tests skipped]

**Refinements**: N passes.

</workflow>