From agents-flow
Audit open GitHub issues for drift against repo state. Flags broken refs, stale claims, and contradictions.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agents-flow:audit-issues [owner/repo | #NN | owner/repo#NN]When to use
Use when auditing a GitHub repo's open issues for drift, broken refs, or stale claims.
[owner/repo | #NN | owner/repo#NN]sonnetThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Audit open issues for drift against repo state. Product: updated issues themselves (mutate on confirm). Read-only by default — mutate only after explicit per-issue confirmation.
Audit open issues for drift against repo state. Product: updated issues themselves (mutate on confirm). Read-only by default — mutate only after explicit per-issue confirmation.
Adopt Skill("orchestrator-rules") for seed-brief contract, CWD verification, and NOTES.md conventions.
Resolve target:
owner/repo from CWD.owner/repo → use directly.#NN → resolve repo from CWD (single-issue audit).owner/repo#NN → split on #.Find local clone at ~/Projects/<repo>. If missing → abort. Fetch latest default branch via git fetch origin (do not check out).
Create .claude/NOTES.md with task list, decisions log, next-action per Skill("orchestrator-rules").
gh issue view <NN> + gh issue list (filtered).gh issue list --state open --limit 0.Read references/detectors.md for detector logic, verdict ranking, and JSON schema — pass pertinent rules into each spawn prompt so workers do not re-read.
Spawn one Agent("agents/workflow-issue-auditor.md") per issue with a <seed-brief> YAML block containing repo, issue_number, cwd, and default_branch_ref. Fan-out in parallel for ≥3 issues; run inline for 1–2.
Each Agent() spawn includes a <seed-brief> YAML block per _shared/seed-brief.md.
See ${CLAUDE_PLUGIN_ROOT}/_shared/composition.md for spawn cost models.
Concatenate JSON reports. Per-issue block: ─── #NN — <title> — verdict: <v> ─── → URL → findings → proposed edit.
Walk blocks in order. Prompt based on verdict:
valid → [s]kipstale/contradicted/unverifiable → [e]dit / [s]kippremise-shifted/superseded → [e]dit / [c]lose / [s]kipMutation rules:
e → Show diff → gh issue edit.c → Post closing comment → gh issue close.s → Advance.Require explicit user approval per mutation.
Require explicit user approval before applying mutations. No unconfirmed mutations.
Read ${CLAUDE_PLUGIN_ROOT}/_shared/compound-on-exit.md. Invoke Skill("compound") exactly once on clean completion.
unverifiable.references/detectors.md before step 4 only.Skill("orchestrator-rules") § Progress tracking.npx claudepluginhub misiekhardcore/agents-flow --plugin agents-flowGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.