coding-agent-quotalisted
Install: claude install-skill a-green-hand-jack/ml-project-repo-agent-native-template
# coding-agent-quota
Use this skill before routing meaningful work between Codex and Claude Code when quota, reset time, provider load, cost-control, or historical task outcomes matter. It covers two layers: (1) quota snapshot + quota-only recommendation (`read_agent_quota.py`, unchanged), and (2) an additive outcome-aware layer (`outcome_route.py` + `outcome_ledger.py`) that compares candidate routes on quota + outcome evidence and falls back conservatively when data is missing or stale.
## Quick Start
Run the bundled script from the repo root:
```bash
python .claude/skills/coding-agent-quota/scripts/read_agent_quota.py --format table
python .claude/skills/coding-agent-quota/scripts/read_agent_quota.py --format json
python .claude/skills/coding-agent-quota/scripts/read_agent_quota.py --role impl --tier 2 --format json
```
Prefer JSON when another agent or orchestrator will consume the result. Prefer table output for human handoff.
## Evidence Rules
- Treat the script output as a snapshot, not a guarantee. Route high-risk or long-running work only after checking `captured_at` and reset times.
- Prefer provider-specific latest valid snapshots; do not require Codex and Claude to come from the same database row.
- If a provider is `unavailable` or stale, say so and route conservatively instead of inventing percentages.
- Do not read or print credential files. The script uses local usage DB/cache/session logs only.
- For subagent launch decisions, pass `--role` and `--tier