cli-cycle

Optimization: This skill uses on-demand loading. Heavy content lives in references/ and is loaded only when needed.

Language rule: Skill instructions are written in English. When generating user-facing output (reports, files, documentation), detect the project's primary language (from README, comments, docs, commit messages) and produce the output in that language. If the project is bilingual, ask the user which language to use before proceeding.

Cycle — Phoenix Continuous Improvement

"The phoenix does not die — it burns what is impure and is reborn stronger. Each cycle consumes the flaws. When there is nothing left to burn, the project is born."

Run every applicable cli-* skill on the current project, collect results, deliver the complete list of corrections in 3 tiers, and loop until clean.

Core Principles

1. You don't duplicate logic — you delegate. Each sub-agent reads the real SKILL.md and follows its instructions. You orchestrate, collect, and judge.
2. Show everything. Never truncate the correction list. The user sees ALL issues, organized by severity.
3. Phoenix loop. Audit → user fixes → re-audit → repeat until convergence.
4. Autonomous convergence. When invoked with --converge, run the entire loop in an isolated worktree, then present a single unified plan. Read references/convergence.md for the full algorithm.
5. Flow graph. Skills form a DAG: audit detects → forge corrects → git commits → re-audit verifies. No cycles allowed. Read references/skill-flow.md for the complete trigger graph, deduplication rules, and cycle detection.

Gotchas — read ../gotchas.md before producing output to avoid known mistakes.

Workflow

Step 0.5 — Determine scope

$ARGUMENTS can be:

• Empty or project root → full project audit (all directories, all skills)
• A subdirectory (e.g., src/features/dlp/) → scoped audit (vertical slice)

Scoped audit rules:

1. Only scan files within the scope directory (and its subdirectories)
2. Pass the scope directory as $ARGUMENTS to every sub-agent: /cli-audit-code {scope}, /cli-audit-test {scope}, etc.
3. Skills that don't accept a directory scope (e.g., cli-forge-pipeline, cli-forge-github) are skipped — they are project-wide by nature
4. The scorecard, triage, and report are titled with the scope: # Cycle — {project} / {scope}
5. Co-located tests are detected within the scope: {scope}/**/test*, {scope}/**/*_test.*, {scope}/**/tests/
6. Co-located docs are detected within the scope: {scope}/**/*.md, {scope}/**/README*