This repository also configures its own agents. See what cc-system tells them →
Agent Claude Code
Turns a vague high-level request into a verifiable spec. Fixes scope, non-goals, and acceptance criteria; records premises that can't be confirmed as assumptions. Stage 1 of the harness-starter orchestrator.
Agent Claude Code
Takes a spec and gathers the context needed for the task (relevant files, conventions, integration points, gotchas) from the codebase into a context-pack. Read-only. Stage 2 of the harness-starter orchestrator.
Agent Claude Code
Takes a single node of the task graph and implements it. Minimal-change principle, verify immediately after each change, work solo. Never touches anything outside the node's scope (file/module). The execution unit of stage 4 of the harness-starter orchestrator.
Agent Claude Code
Takes a spec and context and decomposes the task into a task graph (DAG) of verifiable nodes. Applies logical decomposition (smallest verifiable unit) and physical decomposition (file/module boundaries for parallel safety) separately. Stage 3 of the harness-starter orchestrator.
Agent Claude Code
Verifies the implementation against the spec's acceptance criteria and the plan's node criteria, producing a PASS/FAIL report. Judges only on actual execution/test/observation evidence. On failure, leaves an instruction for which node to fix and how. Stage 5 of the harness-starter orchestrator.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: