Repository navigation
Development
The repository's map is CODEMAP.md and its rules are in AGENTS.md.
python -m unittest discover -s tests # tests
python scripts/cine.py kb lint # schemas, links, copies, manifests
python scripts/cine.py budget # size budgets (green, yellow, red)
python scripts/cine.py build # copy shared/ into skills, rebuild indexes
python scripts/cine.py zip cinewright-continuity # claude.ai upload ZIP in dist/
python scripts/cine.py kb due # entries past their re-check intervalAt v0.3.0 the suite runs 140 tests:
Ran 140 tests in 95.859s
OK
Vocabulary, schemas and the runtime live in shared/ and are copied into each skill by build, with a copied from ... sha256: header. Edit shared/, never the copies; kb lint fails on drift.
budget measures what an agent loads: SKILL.md lines and tokens, description lengths, entry and model card sizes, the runtime, files per plugin, the ZIP. Each measure is green, yellow or red. At v0.2.0 the overall verdict was:
budget: YELLOW (tokens are bytes / 4, an estimate)
evals/run_evals.py runs every skill's evals/evals.json headless (plan, run, report, regrade). Results stay outside the repository.
A release carries one version across .claude-plugin/marketplace.json and the three plugins, a vX.Y.Z tag, and one claude.ai ZIP per skill (built with cine.py zip) attached to the GitHub release.
This wiki describes cinewright 0.3.0 and was last updated on 2026-11-05. The plugin is MIT licensed. Report problems in the issues.
Using it
The releases
Contributing
Elsewhere