Simplify apply

Use this skill

This skill applies one exact removal set already produced by tailrocks-simplify-audit and explicitly approved by the user. It keeps public behavior and necessary validation during each removal. It discovers no candidate, broadens no scope, fixes no defect, and changes no behavior. Naming an audit report grants no mutation authority.

Use this skill when the user passes explicit apply with the exact approved findings. Do not use this skill to audit. Refuse mixed selectors. Never silently invoke the audit owner.

Before you start

Use this skill only when the user explicitly requests it. A model must not select it from task similarity.

Obey the active user request first. If the request conflicts with a safety rule in this skill, stop. Report the conflict.

Before any action, read references/runtime-trust.md and references/behavior-preservation.md. Resolve each relative link against the directory that holds this SKILL.md file.

Run baseline and proof commands only with explicit execution authority and every control below:

• Frozen existing dependencies and inputs
• Scrubbed secrets
• Disabled network
• Owner-only external cache and output
• Bounded time, retries, output, and process tree
• TERM-then-KILL cleanup
• Repository hashes before and after each read-only run.

A red, vacuous, unavailable, or mutation-prone baseline blocks application.

Procedure

1. Bind the approval. Record each item below:

   • The canonical repository
   • The HEAD
   • The merge base and range
   • The dirty state
   • Exact approved finding IDs
   • Touched and dead-code reach
   • Approved characterization-test paths
   • Expected counted deltas
   • Hashes of every writable file.

   Reject stale, ambiguous, symlinked, escaping, behavior-changing, and extra-scope work.

2. Require a preservation oracle. Require relevant existing tests that enter every removed path. Without them, require an explicitly approved characterization test that first passes against unchanged production bytes. Permit reasoned equivalence only for a locally obvious unreachable symbol with complete reference proof. Never permit it for errors, effects, ordering, timing, resources, interfaces, concurrency, authorization, validation, or security.

3. Publish characterization tests first. If approved, publish each characterization-test file sequentially by compare-and-swap. Record its receipt. Prove it against unchanged production bytes.

4. Apply one removal at a time. Apply exactly one approved removal. Publish every production file sequentially by expected-preimage-to-owned-postimage compare-and-swap. Record one receipt per path. Run its focused oracle. Continue one removal at a time. Never claim multi-file atomicity.

5. Roll back rather than repair. On any failed oracle, restore a preimage only by compare-and-swap. Restore only after current bytes prove that they still match the owned postimage. Never use broad Git restoration, repair behavior, or overwrite a concurrent replacement. When full rollback is impossible, retain and name every surviving changed path and recovery artifact.

6. Prove the result. Run focused and full repository gates under the same bounded command contract. Recount every promised measure. Compare the final changed-path allowlist, observable contracts, and repository state with the approved receipt.

Result

The terminal shows APPLIED, BLOCKED, ROLLED_BACK, or RECOVERY_REQUIRED with target identity, approved IDs, pre and post hashes, and tests established before production edits. It shows command and unit receipts, per-removal and total measured deltas, exact changed paths, partial mutations, and recovery artifacts. ROLLED_BACK means every owned mutation was restored and no changed path survives. Any surviving mutation is RECOVERY_REQUIRED. No unapproved edit, behavior change, commit, push, merge, issue, comment, dependency install, or external action occurred.

Completion checks

Before the result is complete, make sure that each item below is true:

• Every applied removal was approved exactly.
• Every observable behavior is preserved.
• Every promised measure was recounted.
• Every changed path has a receipt.
• Partial mutations and recovery artifacts are reported exactly.

References

Read these references at the stated times:

• Read references/runtime-trust.md before any action for the trust rules.
• Read references/behavior-preservation.md before any action for observable behavior and proof levels.