Thanks for helping. This repo is plain Markdown: editing prose is the work. Read
AGENTS.md first — it holds the authoring conventions every change is judged by.
- Fixes: contradictions, broken cross-references, wrong or outdated harness facts (with evidence).
- A new or re-verified harness adapter, made with the
harness-onboardskill — every manifest row carries version, date and how it was verified. - A bundle catalog row, with its verdict added to
docs/repo-survey.md. - Workflow improvements that help most users of most stacks.
- Features that are too specialised or personalised — one person's stack, tools, preferences,
machine or company process. They belong in your fork, your
AGENTS.mdRULES, memory, or a domain master built withdomain-init. - Additions to the always-loaded
workspace_TEMPLATE/AGENTS.mdwithout a strong case: every token there is paid in every session by every user. Default to a skill. - Anything naming a harness outside
adapters/.
- Fork, branch
<type>/<short-slug>, one topic per PR. Commits:<type>: <subject>—featfixdocsrefactorchore. - Change
workspace_TEMPLATE/oradapters/, never an installed copy; try it withworkspace-syncon your own machine. - In the PR, say what changed and how you checked it: terms grepped, cross-references resolve,
frontmatter well-formed,
README.mdstill matches the template.
On Claude Code the repo's own skills need a link, since .claude/ is gitignored:
ln -s ../../.agents/skills/{name} .claude/skills/{name} (Windows:
mklink /J .claude\skills\{name} .agents\skills\{name}).
Contributions are licensed under the repo's MIT License.