Skills Library

VoiceLayer sessions for debriefs, practice, QA, capture, KG review. Triggers: meeting debrief, voice drilling.

Build talks with Workshop/Oren/Uri methods. Triggers: talk prep, slides, pitch, practice, structure review.

Full PR loop: branch, test, commit, push, PR, review, fix, merge — plus the agent-identity signature required on every GitHub write. Triggers: create PR, finish work, post PR comment, sign a comment, golem-id, Co-Authored-By trailer.

Review changes and handle CodeRabbit/Greptile/Bugbot/GitHub feedback. Triggers: review, comments, security.

Life admin for health, habits, jobs, clients, contracts, and scheduling. Triggers: daily plan, wearable data, outreach.

Figma-to-code pixel verification loop. Triggers: Figma designs, pixel perfect, design verification.

Create JSON PRDs for Ralph stories. Triggers: plan feature, create PRD, Ralph stories. NOT running Ralph.

Spawn AI workers in cmux via MCP/repoGolem. Triggers: visible workers, terminal agents, orchestration, audits.

Control cmux panes, splits, browser/sidebar, messages. Triggers: cmux, split panes, terminal reads. NOT shell.

Manage 1Password op secrets/env. Triggers: API keys, .env, op auth.

First-time golems install: check CLIs, wire MCP, symlink skills. Triggers: set up/install golems, new machine.

Arm or stop durable tag-scoped collab-file watches. Triggers: collab monitor, watch collab, listen-name, background watch. NOT for file-integrity auditing or worker-registry completion.

Use when building audio narration dashboards that require STT-after-TTS exact word-timing, a real word-click-seek read-along dashboard, the AfterCode workflow, or publish-to-tailnet delivery.

Create/edit/audit/eval golem skills. Triggers: create skill, skill eval, live eval, JSONL mining.

Use when a user asks to mine a tech or security conference, Hackers Summer Camp, BSides, Black Hat, DEF CON, conference sponsors, exhibitors, or speakers for hiring targets and matching jobs. NOT for general job search or LinkedIn post drafting.

Enforce Cursor=gather, Codex=implement, Claude=orchestrate. Triggers: delegate, worker assignment, routing.

Compare Claude/Codex/Cursor/Wispr changelogs to local setup. Triggers: whats new, changelog, updates.

Security audit for MCP, TS, Swift, shell. Triggers: security, vulnerability, hardening, traversal, injection.

Verified headless Codex worktree fan-out. Triggers: codex-workflows, headless Codex, Codex parallel. NOT for healthy cmux single-worker tasks.

Discipline for cron/loop/monitor payloads. Triggers: cron, /loop, recurring tick, merge queue, no-progress.

Safety gate for destructive git/main commits. Triggers: force-push, reset, branch delete, clean, main commit.

Bulk LLM-judging protocol for fleet-dispatched verdict runs (KG cluster, eval harness). Triggers: judge fleet, bulk judge, R3 verdicts, kg-judge, RT gate, evidence_degraded. NOT for single reviews, Phoenix UX, or non-judge evals.

Mine Wispr SQLite for ASR vocabulary gaps/corrections. Triggers: wispr mining, ASR errors, voice vocabulary.

Fresh-machine golems setup: prereqs, config, repos, MCP, BrainLayer. Triggers: setup, wizard, install golems.

Inspect fleet-authored GitHub PR queues and enforce the one-shot Claude Stop gate. Triggers: PR queue, open-PR blindness, pr-queue gate. NOT for cmuxlayer status-claim enforcement.

Verify deployed infra after merge/restart. Triggers: PR merged, deployed, daemon/MCP restart, live probe.

Push dependency bugs upstream. Triggers: dependency bug, workaround, monkey-patch, fork/fix. NOT app bugs.

Use when code may have duplicated ownership, uncoordinated implementations, multiple writers, split lifecycle controls, live-copy drift, or missing single sources of truth. NOT for generic lint, style, security, or architectural research audits.

Use when auditing cmux panes for claim-or-close, sprint close-out, idle workers, dead shells, uncommitted or unpushed artifacts, untitled seats, or stale numeric surface refs.

Route fan-out / parallel work to the RIGHT engine: Cursor /multitask, headless cursor-agent, Claude Workflow, or cmux. Triggers: multitask, /multitask, parallel agents, fan out, in parallel, batch classify/audit. NOT for one edit or dependent steps.

Bash security checklist: injection, set -euo, printf, quoting. Triggers: shell hardening, bash security.

macOS systems specialist — AppKit NSPanel architecture, launchd services, socket activation, MCP bridge resilience, syspolicyd, high-frequency SwiftUI dashboards. Triggers: menu-bar apps, LaunchAgents, Gatekeeper/TCC, UDS/MCP bridges, 10Hz+ SwiftUI.

Validate assumptions in multi-agent plans; rewrite weak plans. Triggers: validate plan, check assumptions.

Create static HTML dashboards from template. Triggers: dashboard, status, digest, compare, what changed.

Refresh local research context and the Drive grounding corpus. Triggers: stale context, R-number.

Use when about to call a change done; after a frontend change; on a multi-client or multi-provider change; when it works on my path but may not elsewhere; when adding a setting, command, keybinding, or provider adapter; or when changing a wire contract. NOT for single-file leaf fixes with no shared consumer, adapter, entry point, state transition, or user-facing surface.

Run and validate a multi-tier anonymized-peer-ranking council. Triggers: peer ranking, anonymized review, adversarial verification, score multiple candidates. NOT for one authored plan/spec with declared judge families (/plan-council).

BrainLayer retrieval benchmark: recall@K, MRR, p@5, placebo. Triggers: agada bench, BL PR, embedder swap.

Mechanical gate for visual approval routing, in-policy PR admin-merge, and incident framing-first doctrine. Triggers: visual/design approval, CI-green in-policy PR, incident/operator response.

Pre-R0 diff of implementation vs SOTA research. Triggers: before R0, architectural audit. NOT per-PR review.

Fallback for failed brain_store: write memory to docs.local for replay. Triggers: null chunk, DB busy, close.

Non-blocking PostToolUse/Stop hook flagging conserve-Claude / weekly-budget / usage-percent / stay-thin / go-light framing in agent-facing output. Full-throttle is the standing rule; there is NO real conservation premise. Triggers: budget lint, conserve Claude, stay thin, usage percent. NOT for real sourced cost reporting or context/token budget estimation.

Write grounded Claude Desktop/Web research prompts. Triggers: research prompt, Claude Desktop, MCP-aware.

Mechanical collab-first routing gate: no send_input coordination, no sleep-poll ticks, decisions carry a recommendation. Triggers: cross-lead coordination, worker monitoring, async decision.

Create app/Remotion demos. Triggers: demo video, walkthrough, showcase.

Audit/fix AI context hygiene vs ~/.golems/config.yaml. Triggers: context check/bloat, MCP/skill/hook hygiene.

Durable surface→session_id index so /orc RESUMES a crashed lead (repoGolem --resume) instead of duplicate-spawning a fresh one. Triggers: crash resume, resume not respawn, lead crashed, R-036, SPAWN_OVER_RESUMABLE green path, session_id lookup.

Run Cursor-workflow gather/recon via AutoCursor primitives. Triggers: cursor-workflow, autocursor gather, cursor gather.

Brain Drive filing discipline: where artifacts go + how to name them. Triggers: Drive/Brain Drive upload, folders, prompts/results, audits, plans, transcripts, dashboards, docs.local artifacts. NOT for Gemini Drive/web research; >100KB use archive.

Validate Expo iOS/Android sync before first EAS build. Triggers: eas build, prebuild, credentials.

Health checks for MCP, BrainLayer, skills, friction. Triggers: health check, whats broken, MCP/BL status.

Mechanical kill-gate: block 'done/green/fixed' without a same-turn live-outcome probe. Triggers: completion claim, pr-loop completion, deploy/render/dashboard 'done'.

Quiet sprint close: stop monitors, send final dashboard/message. Triggers: fleet wrap, stand down.

Mechanical kill-gate: at a fleet-wrap / stand-down state, assert cron-count==0 — no health-watch / `/loop` / sleep-poll cron left armed. Triggers: fleet wrap, stand down, sprint close, going silent.

Store genuine user corrections/frustration after validating context and speaker intent. Triggers: no/wrong/stop, I told you, not that, profanity/frustration.

NotebookLM/Gemini research: sources, Deep Research, queries, artifacts. Triggers: Gemini, synthesis.

Archive >100KB research/media to Brain Drive. Triggers: Drive upload, archive this, post-digest archival.

Mechanical autonomy gate: flag no-input-decisions + idle-seat-with-open-queue over agent transcripts. Triggers: turn-end autonomy check, /orc + /pr-loop completion, lead about to pause.

Lint a macOS onboarding runbook for privilege-correctness before it ships. Triggers: install runbook, fresh-Mac onboarding, brew cask runbook, admin/Standard user split, happy-camper onboarding.

Lint launchd plists for hardcoded secrets in EnvironmentVariables. Triggers: launchd plist, LaunchAgents, plist secret, raw API key in plist, secrets hygiene, op:// reference.

LinkedIn coach using Aviv Levi 2026 rules. Triggers: LinkedIn posts, ideas, draft review, algorithm fit.

PreToolUse model-pin guard for Fable seats: Agent/Task spawns must pin a non-Fable model; Workflow agent() pins are advisory. Triggers: model pin, Fable seat, Agent spawn, Workflow spawn.

Mechanical monitor-law gate: assert a persistent monitor is armed on the ACTIVE collab channel with a heartbeat marker. Triggers: lead/orc sprint, spawn workers, 'back to silent'.

End-of-day client/work journal. Triggers: nightly journal, daily sweep, check comms, what happened today.

Orchestrate multi-agent sprints/cmux/ecosystem work. Triggers: sprint, spawn, status, catch me up, delegate.

Phoenix human-eval UX: replay, IDs, filters, mark-wrong, mobile, attribution. Triggers: phoenix view, eval UI.

Cross-family judge council over ONE authored plan or spec: 3 visible seats from 3 model families, self-invented rubrics, live-repo read-only validation, measured same-family bias, lift round, exact-head review loops. Triggers: council, review my plan, score the plan, judge the plan, cross-family judges, lift round, exact-head review, conditional GO, MUST_FIX. NOT for anonymized candidate ranking (/adversarial-council) or bulk verdict fleets (/judge-fleet).

Mechanical kill-gate: enforce tri-state QA PASS/FAIL/INCONCLUSIVE — FAIL only for confirmed-observed failures, INCONCLUSIVE when the path couldn't be reached, and a qa-report.md artifact must exist. Triggers: QA verdict, QA done/complete, pre-merge QA, false-FAIL.

Video QA/extraction for screen, local video, YouTube. Triggers: narrated QA, bugs, QA checklist, video gems.

Mechanical kill-gate: block a narration/AfterCode 'render done / give it a play' without a same-turn ls+ffprobe(size>0,duration>0) of the CLAIMED mp3 + a reachable surface + a registered cloned voice. Triggers: render complete, give it a play, narration done, audio mp3 done.

Launch repoGolem agents in any repo. Triggers: spawn agent, launcher, brainlayerClaude, flags -s/-c/-w/--worker.

Write handoff, spawn successor, answer grill, verify orientation. Triggers: hand off, context high.

Deterministic gate for spawn/dispatch briefs that reference spec or handoff files. Triggers: spawn hygiene, assert-before-spawn, handoff brief, spec path preflight.

Use Google Stitch for design-to-code prototypes/tokens, then implement. Triggers: Stitch, screen prototypes.

Mechanical kill-gate: block a dashboard 'published/live' claim unless the SAME turn mirrored it to the tailnet hub AND got HTTP 200. Triggers: dashboard Write, publish to hub, dashboard 'done/live'.

PreToolUse Write/Edit hook enforcing TDD edit limits for implementation files. Triggers: tdd guard, test-first hook, untested implementation block, snapshot/golden carve-out.

Fail-CLOSED PreToolUse guard denying durable-content writes to temp paths (/tmp, /private/tmp, /var/folders, $TMPDIR) across Write/Edit and Bash writes, and enforcing the ratified <repo>/.worktrees/<name> worktree location. Triggers: tmp guard, TMP-BLOCK deny, WORKTREE-CONVENTION deny, bypass ledger. NOT for storage policy.

Deterministic gate for ultracode/comprehensive workflow topology depth floors. Triggers: ultracode, exhaustive audit, deep fan-out, workflow topology, persistent collab.

Orchestrator convergence: mine JSONLs, cite findings, route actions. Triggers: weave, run weave.

Scaffold multi-phase plans with async agents. Triggers: large feature, multi-PR refactor, parallel cmux.

Evidence gate: read files, run tests, verify outputs before claims. Triggers: results, green/complete claims.

List golem skills/descriptions/grades/status. Triggers: list/search/show skills. NOT invoking skills.

Plan sprint transitions, archive work, reset PRD. Triggers: archive, cleanup, sprint transition, reset PRD.

Audit unfamiliar repos: architecture, features, config gaps. Triggers: unfamiliar repo audit. NOT catchup.

Access iCloud Obsidian notes/vault. Triggers: vault, diary, ideas, Hebrew notes, recent notes. NOT BrainLayer.