AI & Agents Agent Skills

Query external AI agents (Codex, Gemini, OpenCode, Claude Code headless) in parallel for independent second opinions, code review, bug investigation, and consensus on high-stakes decisions. Agents and models are configurable in config.json. Use for architecture choices, security review, or ambiguous problems where independent perspectives matter. Not for simple questions answerable from docs or the codebase — use web search or repo exploration instead.

Audits repositories for Claude Code readiness and suggests improvements. Use when asked to check CLAUDE.md quality, review settings, audit project organization, or optimize for agentic work.

Explore and analyze any repository (local path or remote GitHub/GitLab URL) by delegating to Claude Code CLI (`claude -p`) in non-interactive mode with read-only access. Use when the user asks to explore, analyze, investigate, or research a repository or codebase. Triggers on "explore repo", "analyze repo", "investigate repo", "research codebase", "what does this repo do", "how does this codebase work", "ask about repo", "codebase question", "explore repository", "what API does this project have", "analyze this GitHub repo", "explore https://github.com/...", or any request to understand a repository's structure, API, architecture, or implementation details. Works with both local paths and remote URLs (GitHub, GitLab, Bitbucket).

Delegate your agent-skill lifecycle: keep skills current, audit and clean up your inventory, and vet skills before you install them. The Skillsmith Agent diagnoses in full for free, proposes a batched fix plan, and changes files only with your per-changeset approval, with one-step undo. Triggers: "ask the Skillsmith Agent", "clean up my skills", "what skills are outdated", "audit my skills", "vet this skill before I install it".

Use when the user wants to refresh klaussy-generated boilerplate (per-agent conventions, skills, settings, hooks) across every scaffolded agent after upgrading klaussy itself. Re-runs the scaffold against the latest templates so this repo picks up new skills, prompt revisions, and convention rules.

Use when setting up, auditing, or slimming project agent instructions; integrating Teamwork artifacts, MCP/CodeGraph policy, AGENTS/CODEX/CURSOR/CLAUDE rules, install readiness, or reusable workflow-rule migration.

Skillsmith is the canonical lifecycle manager for agent skills (SKILL.md format) across any MCP-capable agent runtime — Claude Code, Cursor, Copilot, Codex, Windsurf. Discover, evaluate, install, use, maintain, author, govern, retire skills. Triggers: 'use Skillsmith', 'ask Skillsmith', 'search Skillsmith', 'find a Skillsmith skill', 'install with Skillsmith', 'Skillsmith trust tier', 'Skillsmith audit', 'create a skill with Skillsmith', 'publish to Skillsmith', 'Skillsmith quota', 'pin a Skillsmith skill', 'compare Skillsmith skills'. Routes natural-language requests to Skillsmith MCP tools and CLI commands.

View and configure settings for coding agents (Claude Code, Codex CLI, OpenCode, and others). Covers JSON settings for Claude Code, TOML for Codex CLI, and JSON/JSONC for OpenCode, including permissions, sandbox, model selection, profiles, feature flags, providers, hooks, subagents, and skills.

Vast.ai CLI to manage GPU instances, volumes, serverless endpoints, and billing.

Create a new Langfuse integration page in the langfuse-docs repo. Use this skill whenever the user wants to add, create, draft, or scaffold an integration page, cookbook, or docs page for a new tool/framework/model-provider/gateway in Langfuse — triggers include "new integration", "integration page", "docs page for <X>", "cookbook for <X>", "add <X> to langfuse docs", or any request that results in a new `cookbook/integration_*.ipynb`. Also use when the user pastes working integration code, a link to a partner's docs, or rough notes and wants them turned into the standard Langfuse integration notebook. The skill produces a correctly formatted Jupyter notebook, updates `cookbook/_routes.json`, and tries to fetch the partner logo into `public/images/integrations/`.

Hands-on playbook for macOS disk cleanup, dev-machine optimization, and proactive health alerting. Use when the Mac is full or slow, when a kernel panic / watchdog timeout / vm-compressor-space-shortage / Jetsam event happened, when the user asks to free disk space, audit storage, set up disk/memory alerts, or restore the same monitoring on a new Mac. Built around Mole (`mo` CLI) for safety guards plus a custom LaunchAgent-based alerter for active warnings. Covers Apple Silicon laptops with heavy AI/Docker workloads. Not for general macOS support, hardware diagnostics, networking issues, GUI / window-manager bugs, Time Machine recovery, broken app installs, or app-specific performance problems unrelated to disk or memory pressure.

Deep research over the Semantic Scholar Graph API. Covers endpoints missing from allenai's lookup skill — paper references (backward citations), recommendations, batch paper lookup (up to 500 IDs), snippet search, and multi-hop citation graph traversal (BFS forward/backward). Use when the user asks to build a citation graph, expand a literature seed, find related work, run a reference network traversal, explore what a paper cites or what cites it beyond simple lookup, or batch-resolve many DOI/arXiv/S2 IDs. For multi-step research questions, delegate to the deep-paper-researcher subagent to keep the main context clean. Not for single paper-by-ID lookups (use semantic-scholar-lookup) or topical discovery (use web_search_advanced_exa).

Orchestrate the a11y accessibility lifecycle: scout → plan → critique → [perspective audit] → test → critique. Dual-mode: full lifecycle automation or step-by-step dispatch. Spawns specialist agents at depth-1 from the main session.

Vast.ai Python SDK — high-level API for GPU instances, volumes, serverless endpoints, and billing.

Splits a large feature branch into smaller, focused pull requests using stacked branches and cherry-pick. Groups commits by concern (infrastructure, application logic, tests, housekeeping), proposes descriptive branch names for user approval, creates stacked branches, and generates What/Why MR titles and descriptions. Use when the user says a PR is too big, asks to split a PR, wants to decompose a branch, or needs to break work into reviewable chunks.

Challenge, push back, play devil's advocate on AI output. Use when: challenge this, are you sure, push back, prove it, what if you're wrong, devil's advocate, stress test, poke holes, second opinion, sanity check, too confident, really?, question this decision. Subcommands: anchor (committed too fast), verify (facts wrong?), framing (wrong problem?), deep (full devil's advocate in separate context).

Safe-to-fail experiment for Complex domain problems where cause-effect is only visible in retrospect. Two-phase: foreground qualify → background probe → sense result. Use when: probe, safe-to-fail, test hypothesis, experiment with hypothesis, Complex domain with hypothesis. NOT for brainstorming (use brainstorm) or known cause-effect (use investigate).

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Begin an explicit Titan service-cohort session with Atlas, Sentinel, and Mneme active and Forge or Delta locked. Use when a Titan service-cohort route needs this explicit bounded step. Do not use for hidden background agents, silent mutation, unreviewed proof sovereignty, or memory canonization without owner confirmation.

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Turn a reviewed diagnosis packet into the smallest honest repair packet with checkpoint posture, rollback markers, health checks, and explicit owner-layer targets instead of silent self-mutation. Use when diagnosis already exists and the next honest move is a bounded repair plan or repair-ready packet. Do not use without a reviewed diagnosis, for playbook-scale rollout work, or for vague self-improvement rhetoric with no bounded target.

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Apply the aoa-source-of-truth-check workflow inside an atm10-* repository using repo-relative public-surface roles, document maps, canonical-file patterns, entrypoint trimming, and local review posture. Use when contributors need a thin project overlay to identify authoritative ATM10 docs or separate active, archived, generated, local-only, and runtime-adjacent surfaces. Do not use when the task is broader policy design, purely code-local, runtime authority, or better handled by the base skill without local adaptation.

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Route OS Abyss ABI, provenance, signatures, SBOM, SLSA/in-toto, Sigstore/Cosign, C2PA, TUF, SCITT, durable evidence, drift, producer profiles, and trust-gate work through existing abyss-machine read models and owner-local producer routes. Use when an agent needs to inspect or consume bundles, containers, models, runtimes, reports, media exports, source seeds, portable memory bundles, browser extensions, generated machine surfaces, or release artifacts without creating a second trust MCP or violating owner boundaries. Do not use for ordinary local edits, raw .aoa session search, or direct signing/promotion/registry mutation without an authorized owner route.

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Classify whether a requested action is safe to proceed, requires explicit approval, or should not be executed. Use when a task may be destructive, security-sensitive, or operationally sensitive and the authority boundary is unclear. Do not use for clearly low-risk work or when the main need is previewing a mutation rather than deciding whether it may proceed.

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Correct, supersede, reindex, or repair graph-coverage drift in AoA decision records by checking graph issues, editing source notes first, rebuilding repo-local indexes, and refreshing the workspace decision graph. Use when a decision note, index metadata, source-surface list, status, supersession, generated index, unknown surface, or graph packet is stale or wrong. When the parent has classified correction but the target decision record, owning source note, repo-local index contract, or graph/owner packet is unavailable inside the active evidence boundary, stop with blocked_missing_input; do not relabel missing input as deferred_owner_boundary. Do not use for pure lookup or for creating a new decision note.

Explicit activation required: do not invoke or load this skill from an implicit match; wait for explicit user or operator invocation or a source-authorized parent-route selection. Create a new AoA decision record in the owning repository by checking graph status/issues, using impact packets for placement, and following repo-local decision law for the actual source note and indexes. Use when a real structural, workflow, tooling, source/export, MCP, skill, or authority decision needs durable rationale. When the parent has classified create but the chosen decision, target repository, source surfaces, graph status, or repo-local decision law is unavailable inside the active evidence boundary, stop with blocked_missing_input; do not relabel missing input as deferred_owner_boundary. Do not use when no decision has been made, the change is trivial, the target repo has unresolved graph issues, or an existing note only needs correction.