geniro-claude-harness

Use when auditing the Geniro plugin repo as a whole — skills, agents, hooks, lib helpers, rules, and docs — for cross-file consistency, stale references, authoring-rule compliance, logic and shell correctness, over-complication, magic numbers, and safety/coverage gaps. Runs a deterministic pre-pass, then parallel dimension reviewers, re-verifies every finding against the cited lines, and writes a tiered report to the local-only design/scratch/. Skip for fixing one known issue (/improve-template) or reviewing a pending code diff (/code-review).

Use when modifying the Geniro plugin itself — fix a Geniro skill, agent, hook, or ARCHITECTURE.md. Researches via parallel agents (codebase + ARCHITECTURE.md + internet), presents evidence, implements after approval. Skip for general codebase Q&A (/geniro:investigate) or app-code bugs (/geniro:debug).

Use when scaffolding a reusable workflow-helper (Slack/PR/release automations) or invoking a previously-created action. Stored at .geniro/actions/. Run-mode executes the action directly — invoking it is the authorization, so no confirmation is asked. Skip for editing core Geniro skills — edit the plugin repo directly.

Use when a bug needs systematic investigation. 3-phase loop (Investigate → Propose → Ship) mirroring /geniro:implement: observe → hypothesize → test → isolate → propose fix → author reproduction test, then escalate to /geniro:implement with a handoff file at .geniro/state/handoff/from-debug-<branch>.md. Adversarial mode authors F→P tests against a diff (verify-changes). Skip for bugs with obvious root cause — go straight to /geniro:implement.

Use when shipping a new feature, endpoint, page, or significant change against a spec.md / plan.md (from /geniro:plan) OR a raw inline task description. 3-phase autonomous loop: Analyze → Implement → Self-review-and-Ship. Optional --deep deepens two phases — a multi-angle self-review with verification escalated only where the call is contested, and a 3× fact-check of the spec's cited claims before editing (higher quality, higher cost).

Use when adding skill-behavior rules at Geniro skill phase boundaries OR cross-cutting code-style rules loaded at every code-writing/review step; also for declaring read-only fact-verification sources (## Data Sources) or routing the agent's memory/learnings through a custom backend like an MCP (## Memory Backend). Operations: list, create, edit, validate, delete. Skip for per-file-pattern rules — .claude/rules/.

Use when starting fresh in an unfamiliar codebase and need rapid orientation. Scans structure and conventions; produces _CODEBASE_MAP.md with architecture, module graph, critical paths, entry points. Skip for specific Q&A (/geniro:investigate) or bug investigation (/geniro:debug).

Use when turning a vague idea or feature request into an approved spec.md before /geniro:implement. Spec-first planning workflow: explore → grill (decision-tree clarification) → propose 2-3 approaches → approve sections → write spec.md → mechanical validate → user approve → handoff. Skip for well-formed specs already authored — use /geniro:implement <path> directly. Optional --deep deepens the analysis — a wider approach search plus a 3-vote majority verification of the spec's cited claims (higher quality, higher cost). Optional --artifact builds a live, auto-updating visual artifact of the plan as it develops.

Use when restructuring code for better organization or reducing tech debt with zero behavior change. 3-phase loop (Plan → Apply → Verify); never ships — the diff is the deliverable. For behavioral changes use /geniro:implement; for performance use /geniro:review (optimizations dimension).

Use when an open pull request has unresolved review comments (human or bot) and/or failing CI checks and you want them triaged into a fix plan rather than fixed by hand. Reads each unresolved thread + failing check, verifies and reproduces it against the code, asks you about the ambiguous ones, then writes a comment-keyed spec.md + a handoff for /geniro:implement — which applies the fixes and, at ship, posts the drafted replies and resolves the threads. Read-only: never edits code or posts to the PR itself. Skip for producing a fresh review of a diff (use /geniro:review) or fixing a located bug with no PR feedback (use /geniro:debug or /geniro:implement).

Use when a comprehensive code review of pending changes (a diff, branch, or PR) is needed. Reporter workflow: triage, a cheap mechanical pre-pass, then parallel single-dimension reviewers (bugs, security, architecture, tests, regressions, conventions, and more, plus any custom ones) whose findings are filtered and individually verified, then persisted. Emits a handoff file at .geniro/state/handoff/from-review-<branch>.md; downstream consumers (/geniro:implement, or you manually) apply the fixes — review never edits code itself. Resolves any needs-your-decision questions before offering the handoff. When the review finds testable bugs, it offers to author failing tests for them (gated by your approval). Optional --deep reviews each check from several angles and majority-verifies findings where the call is contested (higher quality, higher cost).

Use when starting on a new codebase or after a major plugin update. Detects tech stack, generates a project-specific CLAUDE.md (stack, commands, conventions, domain), and validates it. Re-run mode runs a migration sweep. Singleton bootstrap.

Use when the status line shows a plugin update is available, or to manually pull the latest Geniro plugin version. Verifies plugin integrity, ensures user-authored .geniro/instructions/ and .geniro/actions/ survived intact, and walks any breaking changes in MIGRATION.md.

Use when answering deep codebase questions that need evidence — repo structure, code behavior, git history, or internet sources. Parallel research agents produce cited answers. Skip for bug fixes (/geniro:debug) or codebase mapping (/geniro:onboard).

Use when post-hoc analyzing saved Claude conversation threads for pipeline errors — wrong subagent tier, missed parallel-spawn, schema-invalid tool args, premature completion, instruction drift, hallucinated tools. Takes a thread path, or with no argument analyzes the last few work-bearing threads across every project and merges their findings with a recurrence count. Auto-detects JSONL (Claude Code session log) or markdown (UI export / paste). 4-phase loop: Parse → Detect (mechanical + LLM-judge) → Filter → Present with per-finding AUQ. Emits a handoff to /improve-template for approved fixes. Skip for live debugging (/geniro:debug) or pending-diff code review (/geniro:review).

Use when the user wants to turn recent session experience into project rules, or asks what should be learned from the last sessions. Mines past Claude Code session transcripts for durable rule and improvement candidates — recurring user corrections, rejected suggestions, repeated friction — and routes approved candidates to CLAUDE.md / .claude/rules/ / .geniro/instructions/ / learnings. Pass a search string to mine the sessions that mention it; empty picks the most recent working sessions. Skip for questions about the codebase itself (/geniro:investigate) or reviewing a pending diff (/geniro:review).

Use when finding past Claude Code conversation threads — across every project and every config dir on this machine — to hand selected ones to /analyze-thread for failure analysis. Scans the projects/ tree of each config dir, keeps threads that did agentic work (edited code, ran a skill, or spawned a subagent — so read-only review/debug/investigate runs are included), tags each edited or read-only, and either lists them grouped by project or, when you pass a query, searches inside the thread bodies (PR number, error string, filename, or a topic phrase) and ranks the matches. Pass --code-only to restrict to code-editing threads. Takes a free-text pick. Skip for analyzing one already-known thread file (call /analyze-thread directly) or live debugging (/geniro:debug).