GitHub refactor findings
Don't call these scripts directly. Use fragility-common's findings-doctor,
findings-match, and findings-upsert. They load this skill when
fragility-common/config.json sets "tracker": "github". Don't hand-roll
gh issue writes for findings either.
Settings live in config.json: the label, the comment cooldown, and the
per-run ticket cap. The repo is fragility-common's repo. Auth is whatever
gh is logged in as. That account needs triage or write access, and it is
the account the issues are filed under.
How a finding maps onto an issue
- Ticket: an issue with the
refactor-findingslabel. Its key is#<n>;--keyalso takes a bare number or the issue URL. - Plan: the current plan, without its frontmatter, sits in the body
between
<!-- fragility-plan:start … -->and<!-- fragility-plan:end -->. A later sighting with a new plan replaces it. The plan goes in when the issue is created, so no open issue is ever missing it. - State: fingerprint, systems, paths, and sightings go in a hidden
<!-- fragility-finding {json} -->block. Plan versions go in<!-- fragility-plans [...] -->. - Fixed: closed as completed. A new plan for it opens a recurrence issue that names the old one, which GitHub cross-links.
- Declined: closed as not planned or duplicate. It is never re-filed.
- Sightings: comments, with the same cooldown rules as Jira.
The scripts rewrite only the marked sections, so people can edit the rest of the body. Don't edit inside the markers by hand.
Open refactor-findings issues are work orders: the factory picks them up
and implements the plan (see factory-run). Close an issue as not planned to
stop that. Plans from a bug report also get needs-member-approval, which a
member removes to release them.