safety-nets

Vendored (MIT) from dotnet/arcade-skills: .NET engineering-infrastructure CI/CD investigation. A ci-investigator orchestrator agent routes across 10 skills (Helix log analysis, Azure DevOps pipeline triage, binlog/crash-dump inspection, codeflow + dependency tracing, Maestro, known-issue history) and 4 MCP servers. Forked only to fix the manifest agents path for Claude Code's ./-prefixed component-path schema; upstream is unchanged otherwise.

10 skills · 1 agent · 4 MCP servers

Evidence-gated maintainability discipline in two layers. (1) A native plugin: four modes — /nihil:raze (root-authority, write-capable transformation of a repo you own; only secret-leak and catastrophic, unrecoverable commands are blocked) plus /nihil:review, /nihil:implement, /nihil:release with confidence scoring, scope control, and release gating — backed by five read-only review agents and PreToolUse/Stop hooks (a discipline aid, not a security boundary: heuristic command matching, fail-open; a secret / API-key brake is active in every mode). (2) A summonable pantheon of five first-principles dynamic workflows (/nihil orchestrator + /nihil-maat review, /nihil-odin research, /nihil-shiva deletion, /nihil-athena restructure), installed into .claude/workflows/ via /nihil:summon.

6 skills · 1 command · 5 agents · hooks

Code-elegance pipeline: 4 scouts, 2 judges, planner, verifier, and a gated implementer, plus a standalone code-simplifier for single-pass diff cleanups. Persistent state with stage gates.

1 skill · 3 commands · 6 agents

Multi-agent workflow orchestration (7 commands + 2 skills: eight-gates, hades): fix, tournament, mega-swarm, deep-think, batch-implement, red-blue-review, baryon-mode. Hades, eight-gates, and red-blue-review use Teams API for reactive collaboration.

2 skills · 7 commands · hooks

Cognitive amplification stack: epistemic hooks, competitive review, commit integrity checks, and deep-thinking agents.

5 commands · 3 agents · hooks

Mutation-minded, behavior-first test quality. 4 agents (architecture reviewer, senior tester judge, expressive verifier improver, branch coverage implementer) + 4 skills. Coverage is a floor; kill power is the signal.

4 skills · 1 command · 4 agents

Automate API security directly in Claude Code with 42Crunch - automatically audit OpenAPI specs, detect vulnerabilities aligned with OWASP API Security risks (including BOLA/BFLA), and apply AI-powered fixes. Designed for AI-assisted development workflows, it provides continuous guardrails through an audit->scan->remediate->validate loop, ensuring APIs meet enterprise security standards before deployment.

5 skills

Guides AI assistants that use the NetSuite AI Connector: tool-selection order, output formatting, NetSuite domain knowledge, multi-subsidiary and currency handling, and SuiteQL safety guardrails. Requires the NetSuite AI Connector Service.

1 skill

Roll Claude Code out to a team with guardrails you can check. Four deterministic stdlib scripts: a settings linter that reads shared, local, user and managed settings together (array merge, deny-ask-allow order) and blocks blanket shell allows, pre-approved destructive commands and committed credentials; an MCP and plugin inventory that blocks anything missing from your approved list or holding a literal secret, and flags unpinned or plain-http servers; a CLAUDE.md layering audit that resolves every layer and @import and flags broken imports, contradictions, secrets and resident size; and a privacy-safe adoption tracker that turns local session files into counts only and classifies the rollout as PILOT, SPREADING, EMBEDDED or STALLED. Ships cs-claude-code-rollout agent, /cs-claude-rollout and /cs-claude-settings-lint commands, 3 references, a baseline managed-settings policy and a four-phase rollout plan.

1 skill · 2 commands · 1 agent

AI safety and content moderation including Constitutional AI, LlamaGuard, NeMo Guardrails, and Prompt Guard. Use when implementing safety filters, content moderation, or prompt injection detection.

98 skills

Keep Claude Code on task with local-first mode, scope, dependency, hash, and delegation guardrails.

2 skills · hooks

Don't let Claude say "done" until the tests actually pass. A Stop hook runs the repository's real test, typecheck and lint commands and blocks the turn until they pass; a PostToolUse hook warns when an edit weakened a test; a skill teaches Claude to verify before claiming completion. No LLM calls, no network.

2 skills · hooks