analyzing-codebases
Detects project languages and monorepo state, runs language-appropriate static analysis (dependency graph, complexity, duplication, semantic patterns), and…
Executes a refactor plan phase-by-phase on a dedicated branch with per-phase commits and mandatory reviewer checkpoints. Use when characterization-tests scaffold is complete and plan has phases ready to execute.
$ npx -y skills add wayne930242/Reflexive-Claude-Code --skill applying-refactors --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/applying-refactorsContext preview
The summary Claude sees to decide when to auto-load this skill.
Executes a refactor plan phase-by-phase on a dedicated branch with per-phase commits and mandatory reviewer checkpoints. Use when characterization-tests scaffold is complete and plan has phases ready to execute.
name: applying-refactors description: Executes a refactor plan phase-by-phase on a dedicated branch with per-phase commits and mandatory reviewer checkpoints. Use when characterization-tests scaffold is complete and plan has phases ready to execute.
**Applying refactors IS executing each phase as a green, bounded-size commit with an adversarial review between phases.**
One branch. One commit per phase. One reviewer pass per phase. Zero silent advances. Zero phase >400 LOC unless explicitly acknowledged.
**Core principle:** Rollback is a first-class path, not an exception.
**Pattern:** Chain **Handoff:** user-confirmation per phase **Next:** `verifying-refactors` (after last phase green)
**Tasks:** 1. Verify git clean 2. Create refactor branch 3. For each phase: a. Execute edits b. Run verification command c. Dispatch refactor-phase-reviewer d. Commit phase e. Checkpoint with user 4. Handoff to verifying-refactors
Run `git status --porcelain`. Non-empty → abort: "Uncommitted changes exist. Commit or stash, then rerun."
Read plan metadata for `branch_name`. Run:
git checkout -b <branch_name>
Already exists → abort: "Branch exists. Delete or rename, then rerun."
For each `Phase N` in plan (ascending):
Apply changes per phase `files` and `description`. Edit only the listed files. Do not touch adjacent files even if tempting.
Run the phase's `verification.command`. Non-zero exit → stop; do not commit; diagnose. Options: rollback phase, adjust plan, abort run.
Invoke `refactor-phase-reviewer` agent with:
Reviewer returns `APPROVED` or `CHANGES_REQUESTED`.
`CHANGES_REQUESTED` → address feedback in-place, rerun verification, re-invoke reviewer. Max 2 retries → escalate to user.
Per `references/phase-commit-protocol.md` format.
Print phase N summary: files changed, LOC, verification result, reviewer outcome. Ask user:
After last phase's user `continue`, print run summary and hand off to `verifying-refactors`.
| Thought | Reality | |---------|---------| | "Test runner is slow — skip verification" | Green at every commit is the contract. No skip. | | "Reviewer is picky, override" | Reviewer enforces the 400 LOC cap and test discipline. Overriding invalidates the plan. | | "Merge plan phases for speed" | Phase boundary = rollback unit. Merging = losing rollback. | | "amend to fix a typo" | Amend post-review = review bypass. New commit instead. |
A Claude Code plugin marketplace for skills-driven Agentic Context Engineering (ACE) — build, analyze, and maintain agent systems with structured workflows.
Repo: wayne930242/Reflexive-Claude-Code
Detects project languages and monorepo state, runs language-appropriate static analysis (dependency graph, complexity, duplication, semantic patterns), and…
Writes AGENTS.md per subproject, archives run artifacts, and suggests rcc handoff conditionally. Use when verifying-refactors passes (PASS or…
Converts a refactor map into a phased plan using parallel-change, branch-by-abstraction, or strangler fig patterns. Use when user has approved the refactor map…
Adds golden/snapshot tests to untested hotspot modules before refactoring. Use when refactor plan marks any phase with…
Validates hard structural rules (no cycles, file/fn line caps, cognitive/cyclomatic complexity) and runs mutation testing on touched modules. Use when…
Validates component-type choices for agent system work, classifying knowledge as CLAUDE.md vs rule vs skill vs agent vs hook and checking for conflicts. Use…