Point the skill at config show for project context
Merged
RO
15h ago+18+18+01234567890123456789-2-2-0123456789
In a live headless-agent eval, 18 of 22 runs ran cat .basecamp/config.json because skills/basecamp/SKILL.md said to check that file. It is only the local layer: it misses repo, global, env and flag values.
- The skill now points at
basecamp config show --json, whoseproject_idis the effective value with its source. - New eval case
skill-evals/cases/project-context.yml: asked to list todos "in this project", the agent runsconfig showbeforetodos list(help calls do not count). The model eval was not run here (no API key locally).
basecamp-doctor and basecamp-connect had no such guidance.
Summary by cubic
Stops the Basecamp skill from telling agents to read .basecamp/config.json, which only reflects the local config layer and misses repo, global, env, and flag values. The skill now runs basecamp config show --json to get the effective project_id with its source.
- Adds
skill-evals/cases/project-context.yml, which requires the agent to runconfig showbeforetodos listwhen asked for todos in this project.