| name | resolve |
| description | Consume the machine self-diagnostic backlog. Lists the open `machine-report` GitHub issues on the machine repo, ranks them by severity then age, fixes the single highest-priority one at its root cause, verifies green, then closes that issue. The act side of /report (which only writes). Machine-repo-only skill — not shipped to other installs. Trigger: "/resolve", "resolve a report", "triage the reports", "fix the top report", "work the reports backlog".
|
| metadata | {"version":"2.0.0","category":"workflow","status":"active","updated":"2026-06-17","tags":"resolve, report, triage, self-fix, backlog, machine-defect, consume, github, issue"} |
/resolve — fix the top machine report issue, then close it
/report is write-only: it opens defect issues on the machine repo. /resolve is
the act side. One invocation works exactly one issue — the highest-priority
open machine-report — fixes the root cause in the machine, proves the tree is
green, and closes that issue. Run it again to take the next one.
This is a machine-repo-only skill. It lives in .claude/skills/resolve/ but is
deliberately not listed in .claude-plugin/plugin.json skills[], so it is not
shipped when the machine plugin is installed elsewhere — it only resolves the
machine's own repo, where the fixes actually land.
Requirements
gh installed and authenticated (gh auth status), with push access to
yesitsfebreeze/machine. If auth fails, stop and tell the operator to gh auth login.
Scope guard
- Works only on open
machine-report issues on yesitsfebreeze/machine.
- Fixes defects in the machine itself — tools, daemons, hooks, skills, agents,
config. Never treats a report as a target-code feature request.
- One issue per run. Do not batch-fix; each defect deserves its own verification and
commit.
- The fix lands on the shared
main tree: hold the hub branch:main claim while
you edit and commit, release after, and stand down if a live peer holds it
(@.claude/shared/main-lock.md).
- Root cause, never a patch (machine law). If an issue only has symptoms, investigate
until the actual cause is found before editing.
How to run
-
List. Pull the open backlog, newest-first with labels:
gh issue list --repo yesitsfebreeze/machine --label machine-report --state open \
--json number,title,labels,createdAt,url --limit 100
If the list is empty, say the backlog is empty and stop.
-
Rank & pick. Order by severity (severity:blocker > major > minor >
annoyance, read from each issue's labels) then oldest createdAt. Take the top
as the default. Read its full body (gh issue view <number> --repo yesitsfebreeze/machine)
and sanity-check breadth of impact and recurrence — if a lower-ranked issue is
clearly more urgent, override and say why in one line.
-
Diagnose. Read the issue's Evidence and Suspected cause. Reproduce or trace the
defect in the actual machine files before changing anything. Query kern for prior
decisions on the affected area first.
-
Fix. Implement the root-cause fix in the machine (the relevant skill/agent/hook/
config/daemon). Keep exactly one clean implementation; remove any obsolete code in
the same change.
-
Verify. Run /gate (or the narrowest check that exercises the fix) and quote the
passing output. No "done" without evidence. If the defect was in a hook/daemon/skill,
exercise that path directly.
-
Close. Close the issue with a one-line root-cause note:
gh issue close <number> --repo yesitsfebreeze/machine -c "Fixed in <commit/branch>: <root cause>."
-
Report back. One-line summary: which issue, the root cause, the fix, gate result,
and how many issues remain in the backlog.
After the fix
- If the fix changed a documented capability, update the single source of truth (the
skill/agent doc or
/.machine), not a copy.
- Record the decision in kern (
mcp__kern__ingest) when the daemon is up — what defect,
what root cause, what changed — so the next session does not re-decide it.
- If working the issue surfaces a new distinct defect, file it with
/report rather
than expanding this run's scope.