Safe Extraction Protocol

SkillFiles & storage

Apply when extracting code from a large monolith file into submodules. Covers barrel re-exports, _internals DI seam proxy patterns, CI invariant allowlist updates, and cross-file test verification. Prevents CI failures, broken imports, and test regressions from code extraction.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Safe Extraction Protocol skill

What this skill tells your AI

The instructions your AI receives, as published by zaxbyhub/opencode-swarm in .opencode/skills/generated/safe-extraction/SKILL.md and read by ahel’s review.

Follow every step in order. Do not skip steps.

When to use this skill

  • A source file exceeds team-agreed size thresholds (this repo uses <2000 lines per file per FR-005) and needs splitting
  • A subsystem (destructive-command, worktree-isolation, etc.) is being extracted to its own file
  • Code is being moved from one module to another without changing behavior

Benefit: Prevents the three most common extraction failure modes:

  1. CI invariant check failures (new file paths not in allowlists)
  2. Broken _internals DI seams (test mocks stop working)
  3. Cross-file test regressions (other test files that consume the module)

Step 0 — Pre-extraction audit

Before moving ANY code, inventory every path-scoped artifact that references the source file:

0a. CI invariant scripts

grep -rn "<source-file-path>" scripts/ .github/workflows/

Check:

  • LEGACY_EXEMPTS arrays (e.g., check-invariants.sh)
  • Path-scoped lint/scan configurations
  • GitHub Actions path filters

0b. Mock allowlists

grep -rn "<source-file-path>" scripts/mock-allowlist.txt

0c. Test file inventory

grep -rln "from.*<source-module>" src/ tests/ --include="*.test.ts"
grep -rln "vi.spyOn.*<source-module>" src/ tests/ --include="*.test.ts"
grep -rln "_internals.*<source-module>" src/ tests/ --include="*.test.ts"

Record EVERY test file that imports or spies on the source module. These must all pass after extraction.

Prefer the imports tool or repo_map action for comprehensive consumer discovery. Grep catches direct string matches but misses require() imports, dynamic import(), and re-exports through intermediate modules. Use grep as a secondary cross-check.

0d. Import graph

Use the imports tool or repo_map to find all consumers of exports from the source file.

Step 1 — Create the extracted module

  1. Move the code block(s) to the new file(s)
  2. Move all supporting types, constants, and helper functions used exclusively by the extracted code
  3. Add necessary imports to the new file (from external dependencies)

Step 2 — Create barrel re-export (if preserving public API)

If consumers import from the original path, keep the original file as a barrel:

// src/hooks/guardrails.ts (barrel — preserves import path)
// Use EXPLICIT named exports, not `export *`, to avoid naming conflicts
// when multiple submodules export symbols with the same name.
export {
  _internals,
  createGuardrailsHooks,
  enforceSpecDriftGate,
} from './guardrails/index';
export {
  buildEffectiveRules,
  checkFileAuthority,
  getGlobMatcher,
} from './guardrails/file-authority';
export {
  createToolBeforeHandler,
  normalizeToolInput,
} from './guardrails/tool-before';
// etc.

Verify: bun run build succeeds. All existing imports still resolve.

Step 3 — Handle _internals DI seams

If the source module exports _internals for test injection:

3a. Direct functions stay in source _internals

Functions that remain in the source file stay as direct entries:

export const _internals = {
  resolveEvidenceTaskId,      // still in this file
  loadPlanJsonOnly,           // still in this file
  // ...
};

3b. Extracted functions need getter/setter proxies

Functions moved to the extracted module need proxy entries so test mocks propagate:

import { _internals as _extractedInternals } from './extracted-module';

export const _internals = {
  resolveEvidenceTaskId,      // direct
  get extractedFn() {
    return _extractedInternals.extractedFn;    // proxy to extracted module
  },
  set extractedFn(v) {
    _extractedInternals.extractedFn = v;       // allow test injection
  },
};

The extracted module must ALSO export its own _internals:

// extracted-module.ts
export const _internals = {
  extractedFn,
  otherExtractedFn,
};

Type annotation: Always include an explicit type annotation on _internals objects to override as const readonly inference. Without it, as const makes properties readonly and test injection (_internals.fn = mockFn) fails at compile time:

// GOOD — explicit type annotation allows mutation
export const _internals: {
  extractedFn: typeof extractedFn;
  otherFn: typeof otherFn;
} = {
  extractedFn,
  otherFn,
};

CRITICAL: The extracted module's own production code must call mockable functions through _internals.fn(...), NOT through the direct function reference. If extractedFn internally calls otherExtractedFn, it must use _internals.otherExtractedFn() — otherwise test mocks set on _internals won't intercept the internal call. This is the same pattern the parent module follows.

Verify: Run ALL test files from Step 0c — not just the one explicitly in scope.

3c. Alternative: Factory parameter pattern (no _internals proxy needed)

When splitting a factory function into handler files (NOT extracting a subsystem with its own _internals), the getter/setter proxy is unnecessary. Instead:

  1. Handler files export factory functions that receive dependencies as parameters
  2. The orchestrator file calls these factories, passing closure-scoped config
  3. The barrel re-exports only the top-level orchestrator API
// guardrails/tool-before.ts — handler file
export function createToolBeforeHandler(cfg: Config, deps: Deps) {
  // Receives all dependencies as parameters — no _internals needed
  return function toolBefore(input: ToolInput) {
    /* handler logic using cfg and deps */
  };
}

// guardrails/index.ts — orchestrator
export function createGuardrailsHooks(config: PluginConfig) {
  const cfg = resolveConfig(config);
  return {
    toolBefore: createToolBeforeHandler(cfg, deps),
    // ...
  };
}

Use this pattern when:

  • Splitting a large factory function into handler files
  • The submodules don't have their own mockable functions
  • All dependencies can be passed as closure parameters

Use the _internals proxy pattern (3b) when:

  • Extracting a subsystem that tests mock independently
  • The source module exports _internals for test injection
  • Mockable functions are moving to the extracted module

Step 4 — Update CI invariant scripts

For EVERY path-scoped artifact found in Step 0a, add the new file paths:

# Example: check-invariants.sh
LEGACY_EXEMPTS=(
  "src/hooks/guardrails.ts"                    # original (still exists as barrel)
  "src/hooks/guardrails/file-authority.ts"     # NEW
  "src/hooks/guardrails/helpers.ts"            # NEW
  "src/hooks/guardrails/index.ts"              # NEW
)

Verify: Run the invariant check script locally if possible (bash on Windows may require WSL).

Step 4.5 — Re-capture SAST baseline

File-path moves alter SAST finding fingerprints, making pre-existing findings appear new. After extraction:

  1. Run sast_scan with capture_baseline: true and the current phase number, including BOTH the old and new file paths in changed_files
  2. Pass baseline_refresh_rationale (e.g. "file relocation only — content unchanged"): a rename always changes the finding's file identity by design, so post-rename recaptures carry novel findings that must satisfy the absorption gate with an audited rationale — this is intended behavior, not a bug
  3. The merge records a who/when/rationale triage entry for every absorbed finding, so relocated pre-existing findings stop failing the gate without silent acceptance

Verify: SAST scan on the new file paths shows only genuinely new findings, not relocated pre-existing ones.

Step 5 — Update documentation

Update any doc references that point to the old monolith for functions that moved:

grep -rn "<source-file-path>" docs/ *.md

Fix references to point to the new submodule location.

Step 6 — Verification checklist

  • bun run build succeeds
  • bun run typecheck succeeds (zero new type errors)
  • biome ci . passes
  • ALL test files from Step 0c pass (not just the one in scope)
  • CI invariant scripts updated for new paths
  • Documentation references updated
  • No runtime behavior changes (pure extraction)

Common mistakes

MistakeWhy it fails
Forgetting to update LEGACY_EXEMPTSCI quality job fails on process.cwd() check for new file paths
Only testing the explicit test fileCross-file regressions in OTHER consuming test files go undetected
Not creating getter/setter proxies for _internalsTest mocks on the parent module don't propagate to extracted functions
Moving helper functions without updating/re-exporting all consumersImport errors in unrelated files that depended on the helper
Updating docs to reference new paths but missing someStale doc references confuse future readers

Relationship to other skills

  • safe-rename: Use when renaming symbols across the codebase. Use safe-extraction when moving code to new files.
  • mock-to-internals-migration: Use when converting test files from mock.module/vi.spyOn to _internals DI seam. May be needed as part of extraction if the source module's _internals changes.
  • subprocess-safety: Relevant if the extracted code calls spawn/spawnSync — ensure the _internals proxy preserves timeout/kill semantics.

Signals

GitHub stars
467
Forks
51
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
safe-extraction
Source
github.com/zaxbyhub/opencode-swarm