Plugin Claude Code
Dynamically match reasoning effort to task complexity: five tiny skills (low/medium/high/xhigh/max) whose descriptions classify each task, so Claude self-selects the right effort level without setting /effort manually.
Hook
Runs when a session starts on startup and clear, executing inject-routing-instruction.sh. From cfitzgerald-pd/effort-router.
Hook
Runs when you submit a prompt, before the agent sees it, executing remind-routing.sh. From cfitzgerald-pd/effort-router.
Hook
Runs when the agent finishes a response, executing log_effort_usage.py via python3. From cfitzgerald-pd/effort-router.
Skill Claude CodeCodex
Run substantial coding tasks at high effort. Use this for multi-file features, debugging when the root cause is unknown, refactors of any size, writing or overhauling test suites, performance tuning, and integrating new libraries or services. Use it whenever a task needs real investigation or restructures existing…
Skill Claude CodeCodex
Run trivial, mechanical tasks at low effort on a fast model to save time and tokens. Use this for typos, renames, one-line or few-line fixes, formatting, comment or string edits, mechanical find-and-replace, version bumps, simple file lookups, and quick factual questions about code you can answer by reading one place.…
Skill Claude CodeCodex
Run the most expensive-to-get-wrong tasks at maximum reasoning effort. Use this for ALL security-sensitive work — reviewing or hardening auth, crypto, session or input handling counts even when the code change is small, because being wrong is expensive. Also use it whenever the user explicitly asks you to think as…
Skill Claude CodeCodex
Run routine, well-scoped coding tasks at medium effort. Use this for single-file changes, writing a small function or test with clear requirements, fixing a bug whose cause is already known, standard CRUD/boilerplate, config changes, and straightforward docs. Use it whenever a task is ordinary day-to-day work that…
Skill Claude CodeCodex
Report effort-router token savings and likely misroutes: tokens saved by downshifting, turns routed too low that flailed and wasted their tokens, and trivial turns that burned high/xhigh/max reasoning — plus tier/day/model/ session breakdowns. Use whenever the user asks about effort-router stats, token savings, wasted…
Skill Claude CodeCodex
Run very hard tasks at extra-high reasoning effort. Use this for architecture and design decisions (including advice-only discussions with no code changes), gnarly bugs (concurrency, state corruption, flaky or heisenbug behavior), large migrations, and whenever the user says the problem is hard. A normal multi-file…