Thanks for considering a contribution.
Symposium is intentionally small. Contributions should make the loop clearer, safer, or easier to install without turning the project into a large framework.
- Keep skills plain Markdown.
- Keep the canonical source in
skills/. - Do not maintain separate Codex and Claude source files.
- Prefer runtime-neutral wording.
- Keep scratch state under
.symposium/scratch. - Do not let the model auto-adopt user decisions.
Every skill should include:
SKILL.mdplan.mdtest_cases.json- Clear triggers
- Clear "Do NOT use when" boundaries
- A
vs ...distinction from nearby skills
Validate JSON:
for f in skills/*/test_cases.json; do jq empty "$f" || exit 1; doneInstall into a local Codex copy:
./scripts/install.sh --target codex --scope projectInstall into a local Claude copy:
./scripts/install.sh --target claude --scope projectGood PRs are small and focused:
- One skill behavior change, or
- One documentation improvement, or
- One installer improvement.
Please include:
- What changed
- Why it matters
- How you tested it
- More realistic
test_cases.jsonexamples - Example Seeds from real workflows
- Better installer support for other agent runtimes
- A small demo recording or terminal transcript
- Translations that preserve the same canonical skill behavior