c4c

c4c lifecycle specialist. Use when a delegated message starts with `to_subagent: c4c-plan-owner` or when the task is to activate an idea, generate or repair `plan.md` and `todo.md`, decide whether a plan is complete, or close the active plan. This role must follow `plan-lifecycle`, use `idea-to-runbook-plan` when producing `plan.md`, and run `c4c-regression-guard` itself before accepting a close.

Lightweight c4c orchestration shell. Use for the direct user-facing agent that decides whether to call the plan owner, an executor, or a reviewer, checks git status before and after delegation, normalizes canonical regression logs, runs supervisor-side validation, and commits completed coherent slices. This role should stay lightweight.

Use when working on c4cll frontend or backend debugging and you need to choose the right command-line flags for parser, sema, HIR, BIR, or codegen investigation. Covers parse-only, parser-debug channels, canonical dumps, HIR dumps, backend-route BIR observation, and practical command recipes for narrowing failures.

c4c review specialist. Use when a delegated message starts with `to_subagent: c4c-reviewer` or when the supervisor needs an independent review of whether the current implementation path is still aligned with the active source idea by comparing the git diff from the relevant active-idea history point to `HEAD`.

c4c implementation specialist. Use when a delegated message starts with `to_subagent: c4c-executor` or when the task is to execute one bounded packet, modify owned files, update canonical `todo.md` progress for that packet, run the matching build plus task-specific test subset, and hand results back without taking over plan lifecycle. This role should embody the execute-plan behavior for c4c.

Manage the c4c planning lifecycle across ideas/open/*.md, ideas/closed/*.md, plan.md, and todo.md. Use when activating a plan from an open idea, executing an active plan, deactivating or switching plans, or closing a completed idea.

Use when an oversized implementation file, subsystem, or tangled interface family should be rebuilt around cleaner seams instead of receiving another local patch. This skill starts by routing idea creation through `c4c-plan-owner`, requiring a fixed four-idea source-idea series before rebuild work proceeds: markdown extraction, markdown review, replacement draft generation, and draft-to-implementation conversion.

Use when the active c4c route is repeatedly colliding with the same hard problem, the first bad fact keeps moving without reducing the owned failure family, or one monolithic testcase is driving execution. This skill tells the supervisor how to split the problem into focused probes, create a new idea under ideas/open/, and switch the active plan into a decomposition-oriented runbook instead of continuing the stuck route.

Convert a design note in ideas/open/*.md into an execution-oriented plan.md runbook for c4c. Use when the user wants to extract an open idea document into plan.md, normalize planning format, or keep plan generation behavior consistent.

Use when working in this repo on C++ code and you need AST-backed symbol queries instead of reading large files by raw text. Covers function signatures, top-level symbols, definition/declaration lookup, direct callers/callees, and type references through the repo-local `c4c-clang-tool` and `c4c-clang-tool-ccdb` binaries.

Enforce c4c regression guardrails using matching CTest before/after logs. Use when a user asks to validate that a patch did not regress tests, to compare `test_before.log` vs `test_after.log`, or when the supervisor needs to generate missing regression logs itself. Mid-task subset runs such as backend-only coverage are allowed when the before and after commands match exactly.