Agent Claude Code
The read-only adversarial critic for the agentware self-healing review gate. Use to audit a SHIPPED diff against a feature's acceptance criteria at the end of a self-extension run — it constructs hostile/novel inputs, tries to REFUTE each claim, and returns STRUCTURED findings. NEVER edits code; it is an independent…
Agent Claude Code
The implementation agent for the agentware loop. Use to execute the next task in a feature plan ( /work/ /plan.md), verify it with the project's own checks, and log progress. This is the agent the ./agentware.sh loop spawns each iteration.
Agent Claude Code
Plans features for the agentware loop — produces high-quality /work/ /plan.md files but NEVER executes them. Use when the user wants to design, scope, or draft a plan before running ./agentware.sh. Hands off to the loop; does not implement.
Command Claude Code
Create a new feature plan for the loop to execute in this agentware workspace.
Hook Claude Code
Runs before the agent uses a tool, executing retrieval-state.sh. From r5rana/agentware.
Hook Claude Code
Runs after a tool call finishes, executing log-tool.sh. From r5rana/agentware.
Hook Claude Code
Runs when a session starts on startup, resume and clear, executing session-start.sh. From r5rana/agentware.
Hook Claude Code
Runs when you submit a prompt, before the agent sees it, executing log-prompt.sh. From r5rana/agentware.
Hook Claude Code
Runs when the agent finishes a response, executing log-stop.sh. From r5rana/agentware.
Hook Claude Code
Runs when a subagent finishes, executing log-subagent.sh. From r5rana/agentware.
Settings file Claude Code
Agent settings declaring 6 hook events (PreToolUse, PostToolUse, SessionStart, UserPromptSubmit) and 24 allowed tools.
Skill Claude CodeCodex
Verify a running backend endpoint actually works by calling it yourself, not by reading the code. When asked to "verify this endpoint", "test the API", "check the route works", "confirm the mutation persisted", "did the request succeed", or whenever a change adds or modifies an HTTP/RPC/GraphQL endpoint, identify the…
Skill Claude CodeCodex
Audit CI/CD pipelines and agent integration points for injection, privilege escalation, secret exfiltration, and untrusted-input flows. When asked to "audit the CI pipeline", "review these GitHub Actions", "check the build for security", "is this workflow safe", "harden the deploy", or before shipping changes to CI…
Skill Claude CodeCodex
Audit a project's dependencies and supply chain for risk. When asked to "audit dependencies", "check for vulnerable packages", "supply-chain audit", "scan for CVEs", "generate an SBOM", or before shipping/upgrading dependencies, scan the dependency graph for known CVEs, verify lockfile and pinned-version integrity…
Skill Claude CodeCodex
Diagnose and fix a broken developer environment BEFORE work starts. When asked to "set up the dev environment", "why won't this build", "fix my environment", "the project won't run", "check my toolchain", "env doctor", "diagnose setup", or when a fresh clone / new machine / CI runner fails before any real task begins…
Skill Claude CodeCodex
Produce bold, distinctive, accessible UI instead of generic AI-slop output. When asked to "design this screen", "build a UI", "make this look good", "style this component", "improve the design", "make it less generic", "pick colors and typography", "lay out this page", or when scaffolding any new frontend surface (web…
Skill Claude CodeCodex
Turn a working diff into clean Conventional Commits, a well-structured pull request, and a changelog entry — WITHOUT ever committing on the user's behalf. When asked to "write a commit message", "commit this", "draft a PR", "open a pull request", "write the PR description", "squash these commits", "write a changelog…
Skill Claude CodeCodex
Run a coding-interview practice session as the interviewer, and keep a permanent per-problem record of it. Use whenever the user pastes a coding problem (a LeetCode/NeetCode link, a problem statement, or a function signature), or says "interview me", "interview help", "learn help", "let's practice", "mock interview"…
Skill Claude CodeCodex
The knowledge base is agentware's long-term memory across sessions. It does NOT live in this repo — it lives in an EXTERNAL directory the operator chose at onboarding. Resolve that directory at runtime.
Skill Claude CodeCodex
Skill "onboarding" from r5rana/agentware, covering onboarding skill — agentware first-run flow, the core design (state this to the user), self-extension clause (always true), the steps and step 1 — welcome and explain.
Skill Claude CodeCodex
Change data, schema, config, or file layout safely — with a backup and a tested rollback before you touch anything. When asked to "run this migration", "alter the schema", "add/drop a column", "migrate the database", "backfill this data", "rename this field", "change the data model", "move these files", "bulk update…
Skill Claude CodeCodex
Whole-repository static application security testing organized by vulnerability class. When asked to "security audit", "find vulnerabilities", "SAST scan", "pentest the code", or before shipping security-sensitive code, run a recon → per-class detection → verification → report flow covering SQLi, XSS, GraphQL abuse…
Skill Claude CodeCodex
Apply security at design time, before code is written. When asked to "design this securely", "add a secure-by-design review", "what are the security requirements", "harden this feature", or when scoping any new endpoint, auth flow, data model, integration, or LLM/agent feature, walk a proactive checklist grounded in…
Skill Claude CodeCodex
Without an explicit promotion procedure, knowledge dies in three places.