claude-commands

Token-efficient second opinion slash command /advice. Extracts decision point + artifact (≤150 lines), then fans out in parallel: (1) Opus subagent reviewer with fallback chain codex→agy→cursor, (2) /research on the decision topic, (3) /secondo multi-model opinion. Use instead of advisor() which ships the full conversation uncached.

Lazy senior-dev mode — the seven-rung ladder that decides whether to write code at all, whether to reuse code that already exists, and whether to prefer stdlib/platform/installed deps over a new dependency. Use before writing any code, every PR diff, every fix, every feature. Source attribution included.

Use when making any "X is enabled" / "feature X works" / "cache works" claim in a running session — requires runtime probe output from the standard harness startup path, NOT a launchd/cron test that explicitly sets the activation env var.

Pre-flight checklist for self-hosted runner failures. Use when investigating low disk, missing runner, or stuck Green Gate on the jleechanorg/worldarchitect.ai fleet. Always verifies host-level container state, not just GitHub API.

When asked to verify whether a Claude Code feature works (especially slash commands, dialogs, pickers, status indicators), spawn a real interactive TUI session in cmux — never use `claude --print "/feature"` as a test, because --print is non-interactive and will always return "isn't available in this environment" regardless of whether the feature actually works.

Read ~/.hermes/agent-orchestrator.yaml BEFORE declaring an --agent <X> value unsupported. Lists plugin shorthands (wafer, minimax, agy).

Use whenever running an AO spawn. Honors exact --agent/--runtime/--project/--claim-pr; verifies session metadata; never silently substitutes a different worker.

Use BEFORE bulk/iterative AO spawns (>3). Hard cap 20 active workers; spawn ≤5 at a time; never use loadavg as the gate.

Run or maintain the cmux approval worker that scans cmux terminal surfaces for approval dialogs, classifies them with `codex exec`, and sends the matching approval key. Use when testing or operating the launchd-based auto-approver, debugging missed prompts, moving the worker, or tuning candidate detection and approval heuristics.

Codex-native AO evolve loop. Run a deterministic local observe/measure cycle, then optionally delegate targeted fixes via /claw using the canonical evolve-loop skill.

Run the Dark Factory DOT pipeline runner against a goal. Slash command: /factory. Implements StrongDM's Attractor pattern as an external Python runner — .dot files are the versioned artifact, sealed holdouts live in a separate repo, every step is recorded to CXDB, and the Healer clusters failures into diagnoses. Use when you want the goal_harness idea executed as a reproducible external pipeline instead of in-Claude subagent dispatch.

Create Excalidraw diagram JSON files that make visual arguments. Use when the user wants to visualize workflows, architectures, or concepts.

Dark Factory spec workflow (/factory-spec, /fs): create a reviewed spec via the spec_gen pipeline, review an existing spec, or display the pipeline node graphs — spec-review pipelines, factory gates, node types, edge conditions, handler mappings. Create mode runs dark-factory --pipeline slim/spec_gen.dot; review and show modes are in-session only.

Use when adding automation scripts. Requires a caller (CI/cron/hook); watchdog scripts need launchd plist template + deploy.sh install step.

Use BEFORE running any automation loop. Requires stating the observable success output (PR, fixed test, push); rejects prediction-only loops.

Diagnose why the automation system is not autonomously driving PRs to green and merge.

Watch one or more AO worker tmux sessions, classify their state (WORKING/IDLE/QUEUED/DEAD/COMPLETED), auto-remediate known failures (trust TUI, queued prompt), and push-notify on stuck sessions. Reuses ao-session-monitor state detection. Use /babysit to start a monitoring loop on a specific worker, all active workers, or workers matching a PR/branch.

Standardize br bead follow-ups for PR reviews and code reviews so each bead includes exact implementation instructions, API/function signatures, and acceptance criteria.

Use when creating a new branch or entering a worktree. Sets upstream immediately via `git branch --set-upstream-to=origin/<branch>` after first checkout.

Use when testing localhost URLs or web UIs. Enforces mcp__playwright-mcp (headless); forbids mcp__claude-in-chrome__* (requires extension).

Use when dispatching work through the Hermes gateway with /claw, especially when the task may resolve slash commands or hand off into AO worker orchestration.

Use when debugging unauthorized/fallback errors. Flags redacted placeholders (e.g. __OPENCLAW_REDACTED__) as invalid tokens; checks env overrides first.

Control cmux tabs, workspaces, and terminal panes via Unix socket. Use when reading terminal output, sending commands to another agent's pane, switching tabs, or monitoring coder progress.

Read and steer another cmux terminal tab through the Unix socket.

Metric-driven PR code quality review (Short variant for small PRs, Long variant for important/architectural/AI-generated PRs). Cyclomatic / cognitive complexity, duplication, coupling/cohesion, package metrics, code smells, lightweight security patterns, AI-import verification gap. Cite file:line evidence and refactor advice. Used by /code-standards and the /code-quality slash command.

Dispatch independent adversarial reviews for ZFC, ZFC leveling, root-cause-first, and thermo without duplicating their standards.

Example/seed/fixture files must use placeholders not real credentials. Pre-commit grep check + enforcement hooks.

Use BEFORE `mv`/`rm -rf` on any directory. Verifies pwd does not match target; session cwd cannot recover once its directory is gone.

Use for PRs scoped to deletion. Net production LOC must be ≤0; PR lifecycle time-boxed at 30 min; document, do not substitute, the deletion.

Run disk usage analysis and cleanup preview on the local Mac. Always validate snapshot coverage before quoting. Never delete without user approval.

Run disk usage diagnostics, growth history logs, and automated cache/temp cleanup on development machines.

General cross-project evidence standards (user scope). Use for the /es skill. Repo-level .claude/skills/evidence-standards.md files add project-specific extensions — always consult both.

Analyze conversation and git history to find gaps where cold reviews (codex, Bugbot, CodeRabbit, /reviewdeep) caught issues that dark-factory in-pipeline reviewer nodes missed. Proposes targeted .dot and runner improvements ranked by G1–G9 gap taxonomy.

Use when fixing broken automation (CI, gates, hooks). A commit is NOT a fix; verify the fix is active post-deploy; restart gateway after config.yaml changes.

Use when editing GitHub Actions in private repos. Requires self-hosted runners; flags ubuntu-latest as policy drift unless explicitly approved.

Diagnose recurring failures as harness gaps and produce durable guardrail fixes.

Use when /harness is invoked or fixing a harness rule violation. Matches fix durability to severity (nit=memory, wrong=CLAUDE.md, silent-sub=hook+commitment).

Use after every /integrate invocation. Mandates /learn before reporting done; integrate without learn is incomplete execution.

Use when installing/modifying any ~/Library/LaunchAgents plist. Requires template in repo that owns the job; @HOME@ placeholders; install-launchagents.sh cleanup.

Capture durable learnings from failures, corrections, repeated mistakes, successful recovery patterns, or direct /learn requests. Use whenever the user invokes /learn, asks to save/remember/capture a lesson, or when a workflow failure should become reusable process knowledge. Always persist to Claude memory, optional mem0, ~/roadmap, beads, and wiki-ingest.