← ClaudeAtlas

coordination-corelisted

Orchestrate multi-agent work with portable task classification and evidence-driven lifecycle envelopes. Use when any coordinator or worker harness must assign, delegate, steer, supervise, stop, review, or hand off work without assuming a particular model, command, session UI, quota meter, or vendor.
wildcard/codex-to-claude-coordinator · ★ 0 · AI & Automation · score 72
Install: claude install-skill wildcard/codex-to-claude-coordinator
# Coordination Core Coordinate delegated work through a portable contract. Keep product commands, model names, permission controls, and status strings in harness adapters. ## Classify Before Starting 1. Collect only observable inputs defined below and in [references/classification.md](references/classification.md). 2. Run `scripts/classify.py` with those inputs when tool access exists. In an instruction-only surface, apply the same rules below verbatim. 3. Treat difficulty, risk, duration, coordination shape, requested capability tier, and reasoning effort as independent outputs. 4. Ask the selected adapter to resolve the requested tier against capabilities it can prove. Record `unavailable` or `unknown`; never infer availability from an unlabeled or shared usage percentage. 5. Reclassify only when a mechanical input changes. Do not relabel work merely because an agent sounds confident or uncertain. Use the version `0.1` rules exactly: - `difficulty`: `hard` if work crosses systems or a prior attempt failed; otherwise `trivial` for at most one file group; otherwise `standard`. - `risk`: `high` for an irreversible or external action; otherwise `elevated` when a plan gate is required; otherwise `low`. - `duration`: map `small`, `medium`, or `large` expected tool calls to `short`, `medium`, or `long`. - `coordination`: `team` for at least two independent subtasks that cross systems; otherwise `fan_out` for at least two; otherwise `single`. - `reque