How tjboudreaux/cc-thinking-skills works
This repository is a portable catalog of structured-reasoning Agent Skills. A user invokes the router when the right framework is unclear; it selects no skill, one skill, or a small set, while users can also invoke a matching leaf skill directly. The evals/ tree is a substantial supporting product branch: dataset preparation feeds evaluation runners, which use model execution and scoring utilities to produce evidence. The graph reflects documented skill behavior and indexed runner dependencies; unsampled evaluation modules are represented only where their paths and roles are supported.
Main entry points: objective.js Study execution · run-routing-boundary.js Routing boundary study · droid.js Droid model execution · run-scientific-method-experiment.js Scientific-method study
Components 24
Skill catalog
- Model router
skills/thinking-model-router/SKILL.md - Reasoning skills
skills
Evaluation studies
- Dataset preparation
evals/datasets - SWE-bench mapping
evals/datasets/swebench-pro.js - Routing evaluation
evals/run-routing.js - Routing boundary study
evals/run-routing-boundary.js - Calibration runner
evals/run-calibration.js - Objective runner
evals/run-objective.js - Pairwise runner
evals/run-pairwise.js - Scientific-method study
evals/run-scientific-method-experiment.js - Other eval studies
evals
Evaluation runtime
- Skill catalog loader
evals/lib/skills.js - Droid model execution
evals/lib/droid.js - Study execution
evals/lib/objective.js - Pairwise judging
evals/lib/judge.js - Result envelopes
evals/lib/result.js - Run I/O
evals/lib/io.js - Statistics
evals/lib/stats.js - Study conditions
evals/lib/conditions.js
Evidence and analysis
- Evidence registry
evals/lib/evidence.js - Evidence registry data
evals/lib/registry.js - Portfolio analysis
evals/lib/portfolio.js - Published audit
analysis
Other components
- Agent user
Connections 30
- Agent user → Model router: invokes when unsure
README.md - Agent user → Reasoning skills: invokes matching skill
README.md - Model router → Reasoning skills: selects framework
README.md - Dataset preparation → SWE-bench mapping: maps issue rows
evals/datasets/ingest-hf.js - Routing evaluation → Skill catalog loader: loads skill catalog
evals/run-routing.js - Routing evaluation → Droid model execution: queries model
evals/run-routing.js - Routing evaluation → Run I/O: writes run output
evals/run-routing.js - Routing boundary study → Skill catalog loader: loads skill catalog
evals/run-routing-boundary.js - Routing boundary study → Droid model execution: queries model
evals/run-routing-boundary.js - Routing boundary study → Study execution: orders study arms
evals/run-routing-boundary.js - Routing boundary study → Result envelopes: hashes artifacts
evals/run-routing-boundary.js - Routing boundary study → Run I/O: writes artifacts
evals/run-routing-boundary.js - Calibration runner → Droid model execution: runs model calls
evals/run-calibration.js - Calibration runner → Run I/O: uses run I/O
evals/run-calibration.js - Objective runner → Study execution: runs objective study
- Pairwise runner → Droid model execution: runs model calls
evals/run-pairwise.js - Pairwise runner → Pairwise judging: uses judge panel
evals/run-pairwise.js - Pairwise runner → Result envelopes: uses result model
evals/run-pairwise.js - Scientific-method study → Droid model execution: runs model calls
evals/run-scientific-method-experiment.js - Scientific-method study → Study execution: analyzes objective
evals/run-scientific-method-experiment.js - Scientific-method study → Result envelopes: uses result model
evals/run-scientific-method-experiment.js - Scientific-method study → Run I/O: reads and writes
evals/run-scientific-method-experiment.js - Other eval studies → Dataset preparation: prepares evaluation data
evals/datasets/ingest-hf.js - Study execution → Result envelopes: creates result envelopes
evals/lib/objective.js - Study execution → Study conditions: uses study conditions
evals/lib/objective.js - Study execution → Statistics: uses statistics
evals/lib/objective.js - Pairwise judging → Droid model execution: requests judge votes
evals/lib/judge.js - Evidence registry → Evidence registry data: uses registry
evals/lib/evidence.js - Published audit → Evidence registry: reports evidence
- Published audit → Portfolio analysis: summarizes portfolio