90 skills exposing 109 read-only verifier tools over the Model Context Protocol. The catalog is regenerated on every change, and every verifier listed here is registered on the hosted server, which carries 130 tools in total — these verifiers plus the generative Layer 2 agents.
90 skills
Turn a one-paragraph brief into a typed, validated plan.
Use this skill when the user wants to audit Claude Skills for architecture improvements — opportunities to decompose monolithic skills into orchestrator-worker patterns and to right-size model selection (downgrade Opu…
2 MCP tools: architecture_audit, score_complexity
Runs a source-tiered research scan for AI engineering topics (agents, RAG, MCP, context engineering, LLM architecture, evals, governance) and returns a research brief with a claims ledger.
Answers where one piece of Claude Code config belongs, by context cost.
Use this skill when the user wants to design a new Claude Skill — to produce a structured design plan from a one-paragraph brief, including folder structure, frontmatter, output schema, eval shape, and position in the…
1 MCP tool: validate_plan
Designs and audits the measurement layer for a physical campaign before artwork is fixed, because attribution decided after artwork is attribution that can no longer be built.
Finds holes in a plan, design doc, or migration doc, researches surviving claims via firecrawl-mcp to tavily-mcp to WebFetch, and emits a ranked roadmap.
1 MCP tool: plan_shape
Use this skill when the user wants to set up, evaluate, or improve a Ralph Wiggum autonomous-loop workflow on a codebase — assess fit, pick between the bash-loop, ralph-wiggum plugin, or ralph-orchestrator backend, de…
1 MCP tool: ralph_assess_fit
Recommends the right container or runtime technology — Docker, Podman, containerd, or WebAssembly/WASI — for a described workload.
Designs unattended scheduled and event-driven automations for Claude's five schedulers, then verifies the result mechanically instead of asserting it works.
2 MCP tools: schedule_check, unattended_prompt_lint
Mines Claude Code session transcripts under the projects directory to surface recurring workflows worth turning into a skill, slash-command, hook, or memory note.
1 MCP tool: prospect_sessions
Use this skill whenever the user needs the verbatim spec from official docs — JSON shape, API signature, or required-vs-optional fields — before implementing a change.
1 MCP tool: validate_spec_output
Designs and validates agentic workflow specs before any code exists, via scripts/validate_workflow_spec.py (WF001-WF012) and scripts/scan_workflow_patterns.py (WS001-WS004).
2 MCP tools: scan_workflow_patterns, validate_workflow_spec
Author and scaffold skills, servers, and rules through the lifecycle.
Use this skill whenever the user asks to create, author, build, scaffold, or design a Claude Skill in any repo other than claude-skill-toolkit itself — to apply the conventions codified upstream (description rules, sc…
Use this skill when the user wants to wire one Claude Skill to reference another in its hand-off section — to add a next-step bridge, update a strategic recommendation, or insert a sister-skill mention.
1 MCP tool: validate_bridge
Use this skill to drive a new Claude Skill build end-to-end through the toolkit lifecycle once an approved design plan exists.
1 MCP tool: pipeline_status
Scaffolds Claude Code configuration for a project that lacks it.
Deep module for building a Nous Research Hermes agent two ways.
1 MCP tool: validate_hermes_agent
Scaffolds and extends MCP servers using the official mcp[cli] SDK or jlowin/fastmcp 2.x.
4 MCP tools: mcp_audit, mcp_create, mcp_retrofit, validate_mcp_server_project
Read-only planner that converts a hand-authored, non-conformant OKF-ish bundle into strict Open Knowledge Format v0.1 via scripts/plan_okf_migration.py (rules CM001-CM007), with scripts/migrate_okf.py applying the plan.
1 MCP tool: plan_okf_migration
Builds a default-deny claims registry for regulated client marketing and wires a deterministic validator into pre-commit and CI, so an unsubstantiated claim cannot reach print.
Use this skill when adding a new numbered rule, tightening an existing one, or retiring one in any validator skill in this toolkit (skill-preflight-check P-rules, skill-ecosystem-audit M/A-rules, skill-quality-grader…
1 MCP tool: rule_completeness_audit
>- Author runnable workflow scripts for Claude Code's Workflow tool — deterministic multi-agent orchestration files that fan out fresh-context subagents under plain JavaScript control flow.
Score one skill against numbered rules instead of vibes.
Tunes up one Claude Skill across three axes — visibility, determinism, and composability — and returns ranked improvement recommendations (not a pass/fail verdict).
1 MCP tool: hygiene_audit
Use this skill whenever the user is about to upload, package, or publish a Claude Skill — to validate it against Claude.ai's parser, fix common failures, produce an upload-ready zip, or publish to the Anthropic Skills…
1 MCP tool: preflight_check
Use this skill when the user wants to grade, score, or audit the content quality of a single Claude Skill, or evaluate how well its existing evals are written.
2 MCP tools: grade_evals, quality_grade
The GO / NO-GO publish gate and the fix loop that feeds it.
Use this skill when a JSON findings file from a toolkit verifier (skill-preflight-check, skill-ecosystem-audit, skill-quality-grader, skill-agent-architecture-advisor, skill-ralph-loop-advisor) needs each finding clas…
1 MCP tool: dogfood_triage
Drives the adversarial code-review loop until a diff is clean or a stopping condition fires, and grades an executor's return report against a plan_shape handoff plan (EX001-EX007).
2 MCP tools: executor_report_check, sweep_status
Runs a pre-publish GO/NO-GO gate over a finished or just-edited Claude Skill by chaining the toolkit's validators into one binary verdict — skill-preflight-check (parser compliance), skill-quality-grader (Q001–Q009 ba…
1 MCP tool: ship_check
Read-only scanners over the whole repo: drift, context, security, registries.
Audits and maintains the governed asset registry — ledger tables recording every skill, wiki page, reference, workflow, and verifier with its owner, version, lifecycle status, visibility, ACL, usage count, and agent b…
Captures deferred work as ready-to-work GitHub issues a future Claude session can pick up cold, and scans a repo for latent work items to file the same way.
1 MCP tool: scan_backlog
Reconciles what a consultant actually sent a client against what the engagement still needs, then builds the call agenda from the gap.
Read-only linter auditing .claude/workflows/*.js against CE001–CE007 via scripts/lint_context.py.
1 MCP tool: context_audit
Governs a large fan-out ingestion of an external documentation corpus into a repository and reconciles what was written against an authoritative inventory via scripts/reconcile_corpus.py (rules CI001-CI008); scripts/b…
1 MCP tool: reconcile_corpus
Audits a decision log so every entry records what the choice gave up, not only what it won.
Walks the model through the six-phase debugging loop — reproduce, minimise, hypothesise, instrument, fix, regression-test — for any bug (third-party libraries, application code, infrastructure, ops).
2 MCP tools: audit_debug_tags, check_reproducer
Read-only verifier that scans markdown documentation for stale path references, ghost skill names, and dead links after a refactor, service removal, or rename.
1 MCP tool: doc_drift
Use this skill when the user wants to audit, map, consolidate, or rationalize a collection of Claude Skills — to find overlapping descriptions, mismatched scopes, broken hand-offs, or missing evals across a skill dire…
1 MCP tool: ecosystem_audit
Read-only linter that checks a fanned-out subagent brief before it is dispatched, via scripts/lint_fleet_brief.py (rules FB001-FB008).
1 MCP tool: lint_fleet_brief
Runs a post-mortem on a Claude Skill after it was used — reads the session transcript, locates where the skill fired, and quantifies the friction that followed (clarification rounds, corrections, tool retries, token c…
1 MCP tool: scan_friction
Read-only guard flagging merge resolutions that silently revert content, via scripts/audit_merge_semantics.py (rules MG001-MG005).
1 MCP tool: audit_merge_semantics
Designs and scaffolds observability tracks for toolkit capabilities (skills, workflows, scripts, MCP tools) so their cost, findings, verdicts, and friction land in one queryable Postgres table for auditing.
1 MCP tool: scan_tracks
Read-only linter that checks an Open Knowledge Format (OKF) bundle for v0.1 conformance via scripts/audit_okf.py (rules OK001–OK008).
1 MCP tool: audit_okf
Materializes a repo's curated knowledge and its Python code into a committed, drift-guarded Open Knowledge Format (OKF) v0.1 bundle via scripts/export_okf.py, kept in sync by scripts/scan_wiki_drift.py (WK001-WK005) a…
5 MCP tools: audit_code_graph, audit_okf_relations, check_okf_exchange, code_graph_query, scan_wiki_drift
Audits an app's server code for paywall and payment enforcement - Next.js routes, server actions, middleware, Express/FastAPI, Stripe/Polar webhooks, Supabase Edge Functions.
1 MCP tool: paywall_enforcement_audit
Read-only linter that finds rule-id namespace collisions across a whole repo, against RN001-RN004 via scripts/audit_rule_namespace.py, and allocates the next safe id with --next.
1 MCP tool: rule_namespace_audit
Use this skill when the user wants to audit Claude Skills for fit with firecrawl-mcp and tavily-mcp — find skills that could benefit from search grounding, detect which skills already use one or both MCPs, and recomme…
1 MCP tool: search_fit_audit
Audits an application tree for leaked and client-exposed credentials, the way an AI-built Next.js, Vite or Expo app leaks them.
1 MCP tool: secret_exposure_audit
Read-only analyzer that aggregates the contradictions a synthesized documentation corpus records about itself, via scripts/map_contradictions.py (rules SX001-SX006).
1 MCP tool: map_contradictions
Read-only linter that re-verifies every numeric literal, quoted string and endpoint path in a doc authored from a fetched source, against a verification ledger, via scripts/audit_source_fidelity.py (rules SF001-SF008).
1 MCP tool: source_fidelity_audit
Inventories what a client already asserts publicly — website, flyer, booking portal, registry records — into a non-outbound fact table where every disagreement between surfaces becomes a question to ask rather than a…
Validates Claude Code sub-agent definitions (.md files under .claude/agents/) against SA001–SA014 — frontmatter shape, activation phrasing, disambiguation, tool scoping, output-contract and obstacle clauses, per-tier…
2 MCP tools: agent_memory_audit, validate_subagents
Audits a Supabase project's supabase/migrations for row-level-security correctness before the push, mirroring the Supabase Security Advisor.
1 MCP tool: supabase_rls_audit
Use this skill when adding a skill or meta-skill to the claude-skill-toolkit, when enrolling a Claude Code plugin, or when releasing an install-mechanism change.
2 MCP tools: release_checklist, toolkit_registry_audit
Watches the external GitHub repos listed in docs/external-repos.md for changes worth adopting and keeps that registry honest.
1 MCP tool: scan_upstream_watch
Read-only linter finding Python test assertions that cannot fail, against VG001-VG003 via scripts/audit_vacuous_guards.py.
1 MCP tool: vacuous_guard_audit
Read-only linter for vendor performance claims in a doc tree (scripts/audit_vendor_claims.py, VC001-VC011).
1 MCP tool: audit_vendor_claims
Audits a repository's vendored copy of a verifier for fidelity and wiring via scripts/check_vendored_parity.py (rules VV001-VV008); scripts/vendor_verifier.py performs the copy.
1 MCP tool: check_vendored_parity
Load-on-demand engineering discipline and deterministic generators.
Consolidated agent-harness knowledge as one deep module.
3 MCP tools: agents_coverage_audit, anthropic_api_audit, harness_lint
Use this skill when auditing a dev-tool or open-source project's brand-as-code tree (docs/brand/ or brand/) for completeness and governance before assets ship or a listing is submitted.
1 MCP tool: brand_audit
Teaches and audits Claude Code's whole configuration directory — the project's .claude/ folder, the root files and, on request, ~/.claude.
1 MCP tool: claude_dir_audit
Consolidated code-review knowledge as one deep module — a simple interface over a rich reference tree.
1 MCP tool: review_readiness_audit
Use this skill when auditing a landing-page or marketing-copy markdown file for conversion-readiness — structure, heading order, CTA presence and phrasing, and required sections.
1 MCP tool: landing_page_audit
Software-development best practices as one deep module — a simple interface over a rich reference tree.
2 MCP tools: practices_audit, simplicity_audit
Deep module that renders publication-ready, deterministic SVG visuals for business documents from a small declarative spec (YAML or JSON) styled by a design-token theme, so document repos never hand-write chart markup.
Consolidated eval and testing-discipline knowledge as one deep module — the teaching layer over shared/eval-layer.md.
3 MCP tools: eval_testing_audit, evolution_ledger_audit, trajectory_conformance
Drives the operator's LinkedIn content engine end-to-end.
2 MCP tools: scan_content, voice_lint
Consolidated LLM and agent security knowledge as one deep module — a simple interface over a rich reference tree.
2 MCP tools: code_pattern_audit, llm_security_audit
MCP-server security, build discipline, and 2026-07-28 spec currency in one deep module, the server-operator counterpart to llm-security-best-practices.
4 MCP tools: mcp_security_audit, mcp_spec_audit, tool_contract_audit, tool_token_audit
Billing-correctness discipline for a metered service as one deep module — does the money add up.
1 MCP tool: metering_audit
Provider-neutral prompt-engineering discipline — clarity, examples, XML, long-context ordering, thinking/effort calibration, agentic prompting — plus an Anthropic annex of per-model deltas.
Measurement hygiene for a prompt-iteration loop as one deep module - is this number trustworthy.
1 MCP tool: prompt_eval_audit
Consolidated Python-automation discipline as one deep module — how to build a scheduled, unattended script that survives runs nobody watches.
1 MCP tool: automation_audit
Repository orientation and legibility as one deep module — audits whether a repo explains itself to a newcomer or an agent.
1 MCP tool: repo_clarity_audit
Retrieval-pipeline discipline as one deep module — does the pipeline find the right passages, and would anyone notice if it stopped.
1 MCP tool: retrieval_audit
AI-native software-delivery discipline as one deep module — does the chain link up, and does each policy have something behind it.
1 MCP tool: sdlc_audit
Use this skill when designing, reviewing, or auditing a website, landing page, or product UI for whether its stated reader can actually read and scan it — type scale and measure, plain-language glossing of jargon, car…
1 MCP tool: ux_audit
Converts a Vue 3 application shell from a horizontal top navigation bar to a vertical left sidebar with a collapsible icon-only mode, a CSS custom-property design-token layer, and consistent card layouts.
2 MCP tools: audit_shell, validate_shell
Use this skill when auditing a CSS stylesheet or CSS custom-property token file for token-discipline and consistency problems before it ships.
1 MCP tool: design_tokens_audit
Consolidated writing and de-slop knowledge as one deep module — a simple interface over a rich reference tree plus a deterministic de-slop linter.
1 MCP tool: deslop_lint
Not yet classified into a lifecycle group.
Read-only auditor for Google Antigravity's agent surfaces, plus a post-mortem on a run that stopped early.
2 MCP tools: agy_stall_audit, agy_surface_audit
Read-only linter for Claude Code surfaces specific to this ecosystem, kept separate from the toolkit's provider-neutral deep modules.
3 MCP tools: claude_md_audit, plugin_manifest_audit, settings_permission_audit
Maintains a plain-English record of what changed in a repository and why, rendered from a machine ledger into docs/DECISION_LOG.md and never hand-edited.
1 MCP tool: change_history_audit
Scans a Claude skill bundle's content for injected instructions, hidden text, fetch-and-execute, encoded execution, unpinned deps, shipped binaries, credential harvesting, persistence, and dynamic code sinks (SK001-SK…
3 MCP tools: check_agentshield_report, check_scan_report, skill_security_audit
Audits a decision-model fine-tune configuration for reproducibility discipline (FT001-FT012), pairing with the calibration-report audit so promotion needs both a pinned run and honest numbers.
Authors, audits and trims the always-loaded instruction layer an assistant reads before it is asked anything — chat-app Project instructions, account profile instructions, CLAUDE.md, .claude/rules files, CLAUDE.local.…
1 MCP tool: instruction_layer_audit
Turns a prose operator procedure — a migration note, a credential-issuance step, a grant script — into a copy-paste runbook whose every credentialed command fetches its secret inline, preceded by preflight checks for…
Authors a prompt from an intent when none exists yet, then hands it to the measured loop rather than claiming it works.
1 MCP tool: prompt_lint
Selects the example prompts a product publishes for its tool surface by measurement instead of authorship, so a documented prompt is proven to reach the tool it claims.
3 MCP tools: prompt_suite_check, prompt_suite_run, prompt_suite_scaffold
Use this skill for a first, thorough harvest evaluation of an external repository — deciding what a toolkit should adopt from it and writing the verdict down durably.
1 MCP tool: analysis_record_audit
Audits a body of artifacts against an external published standard, and schedules that audit to repeat unattended.
