rescue

Purpose

Legacy refactoring orchestrator for safely modernizing messy codebases. Rescue runs a multi-session workflow: assess damage (autopsy), build safety nets (safeguard), perform incremental surgery (surgeon), and track progress (journal). Designed to handle the chaos of real-world legacy code without breaking everything.

Triggers

• /rune rescue — manual invocation on legacy project
• Auto-trigger: when autopsy health score < 40/100

Calls (outbound)

• autopsy (L2): Phase 0 RECON — full codebase health assessment
• safeguard (L2): Phase 1 SAFETY NET — characterization tests and protective measures
• surgeon (L2): Phase 2-N SURGERY — incremental refactoring (1 module per session)
• retro (L2): post-rescue retrospective — capture lessons learned
• journal (L3): state tracking across rescue sessions
• plan (L2): create refactoring plan based on autopsy findings
• review (L2): verify each surgery phase
• session-bridge (L3): save rescue state between sessions
• onboard (L2): generate context for unfamiliar legacy project
• dependency-doctor (L3): audit dependencies in legacy project
• context-pack (L3): create structured handoff briefings before spawning subagents
• neural-memory | Phase start + phase end | Recall past refactoring patterns, capture new ones

Called By (inbound)

• User: /rune rescue direct invocation
• team (L1): when team delegates rescue work

---

Execution

Step 0 — Initialize TodoWrite

Rescue is multi-session. On first invocation, build full todo list. On resume, read RESCUE-STATE.md and restore todo list to current phase.

TodoWrite([
  { content: "RECON: Run autopsy, onboard, and save initial state", status: "pending", activeForm: "Assessing codebase health" },
  { content: "SAFETY NET: Add characterization tests and rollback points", status: "pending", activeForm: "Building safety net" },
  { content: "SURGERY [Module N]: Refactor one module with surgeon", status: "pending", activeForm: "Performing surgery on module N" },
  { content: "CLEANUP: Remove @legacy and @bridge markers", status: "pending", activeForm: "Cleaning up markers" },
  { content: "VERIFY: Run full test suite and compare health scores", status: "pending", activeForm: "Verifying rescue outcome" }
])

Note: SURGERY todos are added dynamically — one per module identified in Phase 0. Each module gets its own todo entry.

---

Phase 0 — RECON

Mark todo[0] in_progress.

Call neural-memory (Recall Mode) for past refactoring patterns in similar codebases.

0a. Full health assessment.

REQUIRED SUB-SKILL: rune:autopsy
→ Invoke `autopsy` with scope: "full".
→ autopsy returns:
    - health_score: number (0-100)
    - modules: list of { name, path, loc, cyclomatic_complexity, test_coverage, health }
    - issues: list of { severity, file, description }
    - recommended_patterns: map of module → refactoring pattern

0b. Generate project context if missing.

Check: does CLAUDE.md exist in project root?
  If NO:
    REQUIRED SUB-SKILL: rune:onboard
    → Invoke `onboard` to generate CLAUDE.md with project conventions.

0c. Audit dependencies.

REQUIRED SUB-SKILL: rune:dependency-doctor
→ Invoke `dependency-doctor` to identify: outdated packages, security vulnerabilities, unused deps.
→ Capture: dependency report (used in surgeon prompts).

0d. Save initial state.

REQUIRED SUB-SKILL: rune:journal
→ Invoke `journal` to write RESCUE-STATE.md with:
    - health_score_baseline: [autopsy score]
    - modules_to_rescue: [ordered list from autopsy, worst-first]
    - current_phase: "RECON complete"
    - sessions_used: 1
    - dependency_report: [summary]

REQUIRED SUB-SKILL: rune:session-bridge
→ Invoke `session-bridge` to snapshot state for cross-session resume.

Bash: git tag rune-rescue-baseline

0e. Build module surgery queue.

From autopsy.modules, filter: health < 60
Sort: ascending health score (worst first)
Add one TodoWrite entry per module:
  { content: "SURGERY [module.name]: [recommended_pattern]", status: "pending", ... }

Mark todo[0] completed.

---

Phase 1 — SAFETY NET

Mark todo[1] in_progress. This phase runs once before any surgery.

1a. Characterization tests.

REQUIRED SUB-SKILL: rune:safeguard
→ Invoke `safeguard` with action: "characterize".
→ safeguard writes tests that capture CURRENT behavior (even buggy behavior).
→ These tests are the rollback oracle — if they break, surgery went wrong.
→ Capture: test file paths, test count.

1b. Add boundary markers.

REQUIRED SUB-SKILL: rune:safeguard
→ Invoke `safeguard` with action: "mark".
→ safeguard adds inline markers to legacy code:
    @legacy     — old implementation to be replaced
    @new-v2     — new implementation being introduced
    @bridge     — compatibility shim between old and new

1c. Config freeze + rollback point.

REQUIRED SUB-SKILL: rune:safeguard
→ Invoke `safeguard` with action: "freeze".
→ safeguard commits current state as checkpoint.

Bash: git add -A && git commit -m "chore: rescue safety net — characterization tests + markers"
Bash: git tag rune-rescue-safety-net

Mark todo[1] completed.

---

Phase 2-N — SURGERY (one module per session)

For each module in the surgery queue (one per session):

Mark the corresponding SURGERY todo in_progress.

Sa. Pre-surgery check.

Verify:
  [ ] Safety net tests pass (run characterization tests)
  [ ] Module is not coupled to another in-progress module
  [ ] Blast radius ≤ 5 files

Blast radius check:
  Bash: grep -r "import.*[module-name]\|require.*[module-name]" --include="*.ts" --include="*.js" -l
  Count files. If > 5:
    → STOP surgery on this module
    → Report: "Blast radius [N] files exceeds limit of 5 — use Strangler Fig pattern to reduce scope first"
    → Pick a smaller sub-module to start with

Sb. Execute surgery.

REQUIRED SUB-SKILL: rune:surgeon
→ Invoke `surgeon` with:
    - module: [module name and path]
    - pattern: [recommended_pattern from autopsy]
    - blast_radius_files: [list from pre-surgery check]
    - dependency_report: [from Phase 0]
    - characterization_tests: [paths from Phase 1]

Supported patterns:
  Strangler Fig          — for modules > 500 LOC: route traffic to new impl gradually
  Branch by Abstraction  — for replacing implementations: introduce interface first
  Expand-Migrate-Contract — for safe transitions: expand API, migrate callers, contract old API
  Extract & Simplify     — for cyclomatic complexity > 10: extract pure functions

surgeon returns: modified files list, refactoring summary, test results.

Sc. Review surgery output.

REQUIRED SUB-SKILL: rune:review
→ Invoke `review` with: modified files, surgeon summary.
→ review checks: code quality, pattern adherence, no regressions introduced.
→ Capture: review verdict (pass | fail | warnings).

If review verdict == fail:
  → STOP, do not commit
  → Report review findings to user
  → Revert surgeon changes: Bash: git checkout [modified-files]

Sd. Run characterization tests.

Bash: [project test command, e.g. npm test or pytest]
If tests fail:
  → STOP immediately
  → Report: "Characterization tests broken by surgery on [module] — reverting"
  → Bash: git checkout [modified-files]
  → Do NOT mark todo complete
  → Update RESCUE-STATE.md with failure note

Se. Commit and save state.

Bash: git add [modified-files]
Bash: git commit -m "refactor([module]): [pattern] — [brief description]"

REQUIRED SUB-SKILL: rune:journal
→ Update RESCUE-STATE.md:
    - module [name]: status=complete, health_before=[X], health_after=[Y]
    - sessions_used: [increment]

REQUIRED SUB-SKILL: rune:session-bridge
→ Save updated state for next session resume.

Context check — before continuing to next module:

If approaching context limit (50+ tool calls or user signals fatigue):
  → STOP after current module commit
  → Report: "Session limit reached. Rescue state saved. Resume with /rune rescue to continue."
  → Do NOT start next module in same session

Mark SURGERY todo completed.

Repeat for each module in queue across subsequent sessions.

---

Phase N+1 — CLEANUP

Mark CLEANUP todo in_progress.

Run only after ALL surgery todos are completed.

Remove boundary markers.

Grep: find all @legacy, @bridge markers in codebase
  Bash: grep -rn "@legacy\|@bridge" --include="*.ts" --include="*.js" -l

For each file with markers:
  → Remove @legacy blocks (old implementation replaced)
  → Remove @bridge shims (migration complete)
  → Keep @new-v2 comments only if they add documentation value; otherwise remove
  Edit each file to strip markers.

Verify markers removed.

Bash: grep -rn "@legacy\|@bridge" --include="*.ts" --include="*.js"
Expected: no output. If any remain → fix before continuing.

Bash: git add -A && git commit -m "chore: rescue cleanup — remove @legacy and @bridge markers"

Mark CLEANUP todo completed.

---

Phase N+2 — VERIFY

Mark VERIFY todo in_progress.

Bash: [full test command]
Capture: passed, failed, coverage %.

If tests fail:
  → Do NOT mark rescue complete
  → Identify which module introduced failure
  → Report: "Final verify failed: [failing test list]"

REQUIRED SUB-SKILL: rune:autopsy
→ Invoke `autopsy` again with scope: "full".
→ Capture: health_score_final.

Compare health scores.

health_score_baseline: [from Phase 0 RESCUE-STATE.md]
health_score_final:    [from this autopsy]
improvement:           [final - baseline]

Report:
  Rescue complete.
  Health: [baseline] → [final] (+[improvement] points)
  Modules refactored: [count]
  Sessions used: [count]

REQUIRED SUB-SKILL: rune:journal
→ Final RESCUE-STATE.md update: status=complete, health_final=[score].

Bash: git tag rune-rescue-complete

Call neural-memory (Capture Mode) to save refactoring patterns and decisions from this rescue.

Mark VERIFY todo completed.

---

Status Command

/rune rescue status — reads RESCUE-STATE.md via journal and presents:

## Rescue Dashboard
- **Health Score**: [before] → [current] (target: [goal])
- **Modules**: [completed]/[total]
- **Current Phase**: [phase]
- **Sessions Used**: [count]

### Module Status
| Module | Status | Health | Pattern |
|--------|--------|--------|---------|
| auth | done | 72→91 | Strangler Fig |
| payments | in-progress | 34→?? | Extract & Simplify |
| legacy-api | pending | 28 | TBD |

---

Safety Rules

NEVER refactor 2 coupled modules in same session
ALWAYS run characterization tests after each surgery
Max blast radius: 5 files per session
If context low → STOP, save state via journal + session-bridge, commit partial
Rollback point: git tag rune-rescue-baseline (set in Phase 0)

Constraints

1. MUST run autopsy diagnostic BEFORE planning any refactoring — understand before changing
2. MUST create safety net (characterization tests via safeguard) BEFORE any code surgery
3. MUST NOT refactor two coupled modules simultaneously — one module per session
4. MUST run full test suite after EVERY individual edit — never accumulate failing tests
5. MUST tag a safe rollback point before starting surgery
6. MUST NOT exceed blast radius of 5 files per surgical session

Mesh Gates

Gate	Requires	If Missing
Autopsy Gate	autopsy report with health score before planning	Run rune:autopsy first
Safety Gate	safeguard characterization tests passing before surgery	Run rune:safeguard first
Surgery Gate	Each edit verified individually (tests pass)	Revert last edit, fix, re-verify

Output Format

## Rescue Report: [Module Name]
- **Status**: complete | partial | blocked
- **Modules Refactored**: [count]
- **Tests Before**: [count] ([pass rate]%)
- **Tests After**: [count] ([pass rate]%)
- **Health Score**: [before] → [after]
- **Rollback Tag**: [git tag name]

Returns

Artifact	Format	Location
Rescue state	Markdown	RESCUE-STATE.md (updated each session)
Characterization tests	Source files	Written by rune:safeguard per module
Refactored modules	Source files	Modified in-place, committed per surgery session
Health score comparison	Inline (Rescue Report)	Baseline vs final autopsy scores
Rescue Report	Markdown (inline)	Emitted at session end (per module and final)

Document Ownership

Scope	Access	Files
Owns (read + write)	RESCUE-STATE.md, .rune/rescue-*.md, git tags (rune-rescue-*), refactored source files (one module per session)
Reads (never writes)	CLAUDE.md, autopsy reports, safeguard test files, .rune/contract.md
Never modifies	Test files written by safeguard (characterization tests are the safety net — rescue reads them, never edits), compiler/**, SKILL.md files

Rescue delegates to surgeon for actual code changes and safeguard for test creation. Rescue owns the orchestration state, not the artifacts.

Anti-Patterns

Common legacy refactoring failures. These turn "rescue" into "make it worse."

Anti-Pattern	Why It Fails	Correct Approach
Big bang refactor — rewriting everything at once	No rollback path. One bug = entire refactor is broken	One module per session. Commit + verify after each module
Surgery without safety net — refactoring before characterization tests exist	No way to verify behavior is preserved. "It compiles" ≠ "it works"	HARD-GATE: safeguard tests must pass BEFORE any surgery begins
Coupled module surgery — refactoring two interdependent modules simultaneously	Changes in module A break module B mid-surgery. Neither is stable	One module per session. Stabilize A completely before touching B
Ignoring the autopsy — starting refactoring without understanding current health	Fixes symptoms not causes. Wastes effort on low-impact modules	Phase 0 autopsy is mandatory. Surgery queue ordered by impact, not convenience
Prototype patterns in production — replacing legacy code with quick-and-dirty rewrites	Creates new legacy. The "rescue" becomes the next rescue target	Apply proven refactoring patterns from autopsy recommendations. Clean code, not fast code
No rollback point — surgery without a tagged baseline	If surgery goes wrong, no safe state to return to	git tag rune-rescue-baseline before ANY code changes

Sharp Edges

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

Failure Mode	Severity	Mitigation
Starting surgery before safety net committed and tagged	CRITICAL	HARD-GATE: rune-rescue-safety-net git tag must exist before Phase 2
Refactoring two coupled modules in the same session	HIGH	HARD-GATE: one module per session — split coupled modules into sequential sessions
Blast radius > 5 files before surgery halted	HIGH	Count importers before each surgery — stop if > 5 and split scope
Not saving state between sessions (rescue spans many sessions)	MEDIUM	journal + session-bridge mandatory after each session — RESCUE-STATE.md must be current
Continuing surgery after characterization tests fail on current code	MEDIUM	Tests must PASS on unmodified code first — fix the test if current behavior is captured wrongly

Done When

• autopsy complete with quantified health score and surgery queue
• safeguard characterization tests passing on current code (HARD-GATE)
• All modules in surgery queue processed (one per session)
• @legacy and @bridge markers removed from codebase (CLEANUP phase)
• Final autopsy run — health_score_final > health_score_baseline
• Rescue Report emitted with before/after health comparison and session count

Cost Profile

~$0.10-0.30 per session. Sonnet for surgery, opus for autopsy. Multi-session workflow.