MCP server for AI coding agents — persistent engineering memory, knowledge compounding, and spec-driven development workflows. Part of TRW Framework.
Skill Claude Code
Run /trw-audit to verify code against PRD acceptance criteria, enforce the mandatory NFR checklist, and assess test quality. Use this skill when an independent, bias-breaking audit is required, invoking the command /trw-audit against a PRD id to generate the adversarial spec-vs-code report.
Skill Claude Code
Load the TRW ceremony reference with the tool lifecycle table, execution phases, and example flows. Use this to resolve questions about tool selection and execution timing within the TRW framework. Invoke: /trw-ceremony-guide.
Skill Claude CodeCodex
Commit requested work using repository-native conventions, scoped changes, and verified evidence. Use only when the user asks for commits, including an ongoing request to commit frequently.
Skill Claude Code
Hand work to other agent CLIs (codex, agy, grok, claude, ...) with one trwdispatch call: review, critique, plan or implement, on one harness or fanned out to several. Use when asked to "get codex/agy/grok to review / critique / plan", or when a second opinion from another model would help.
Skill Claude CodeCodex
Persist the session's work with validation evidence, durable learnings, and client instruction sync. Use: /trw-deliver.
Skill Claude Code
Scan files for duplicated code blocks and classify consolidation candidates. Read-only analysis — reports evidence without modifying code. Use: /trw-dry-check [file-patterns...].
Skill Claude Code
Convert an approved PRD into a repository-grounded execution plan with behavior-sized tasks, verified paths/interfaces, tests or other proof, dependencies, ownership, migration concerns, and exact project-native verification. Internal phase used by trw-prd-ready and trw-prd-ready.
Skill Claude Code
File a TRW bug, install issue, feature request, feedback note, or question through the authenticated POST /v1/submissions channel. Prompts for the category and a one-line summary, optionally attaches recent learnings + the last error trace, and returns a submissionid the operator can quote on follow-up. Use…
Skill Claude Code
Check TRW framework compliance. Reports ceremony adherence, phase gate status, and active run health. Use when unsure if framework obligations are being met. Use: /trw-framework-check.
Skill Claude Code
Record, update, or retire learnings. With a summary: records a new learning. With "resolve/obsolete L-id": changes status. Without arguments: reflects on session memory. Use: /trw-learn ["summary" | resolve L-id | obsolete L-id].
Skill Claude Code
Audit learning memory health. Shows tag distribution, impact spread, staleness, duplicate candidates, and recommendations. Read-only. Use: /trw-memory-audit.
Skill Claude Code
Optimize learning memory. Prunes stale entries, consolidates duplicates, rebalances tags. Interactive — confirms before deleting. Use: /trw-memory-optimize.
Skill Claude Code
Refine Product Requirement Documents (PRDs) to sprint-ready quality by researching the codebase, drafting missing sections, and iterating until validation passes. This skill is triggered exclusively by the /trw-prd-ready or /trw-prd-ready commands. Do not invoke directly by users.
Skill Claude Code
Prepare new requirements and execution plan together before review; preserve existing PRD routes. Accepts a feature description ("Add rate limiting") or a PRD ID (PRD-CORE-EXAMPLE). Use: /trw-prd-ready "Add rate limiting to the API" or /trw-prd-ready PRD-CORE-EXAMPLE.
Skill Claude Code
Assess PRD quality and output a structured READY/NEEDS WORK/BLOCK verdict with per-dimension scores. The skill is read-only and never modifies files. Trigger only via automatic invocation from /trw-prd-ready or /trw-prd-ready; do not invoke directly.
Skill Claude Code
Audit TRW health in the current or target project. Summarizes tool usage, ceremony compliance, hook enforcement, active runs, learnings, and issues. Use: /trw-project-health [path].
Skill Claude Code
End-of-session reflection: examine the session's actual signals (events, diffs, build results, corrections), identify process/organization/tooling/agent improvement opportunities, get approval, then route each one to implementation (inline quick-fix, PRD pipeline, learning, or backlog) with a follow-through ledger.…
Skill Claude Code
OWASP-focused, language-aware security audit of a target repository. Checks command injection, unsafe deserialization, path traversal, secrets, dependency risk, and input validation. Read-only. Use: /trw-security-check [module or 'all'].
Skill Claude Code
Audit test coverage and strategy. Identifies untested modules, coverage gaps, and suggests test improvements. Use before or during IMPLEMENT phase. Use: /trw-test-strategy [module or 'all'].
Skill Claude Code
Run a focused pre-audit review of modified functionality and tests. Verify requirement evidence, wiring, applicable NFRs, test quality, and changed-slice simplicity before requesting an independent audit.
Skill Claude CodeCodex
Propose a work plan to a formation peer, review a peer's plan, and tell a current review from a stale one. Use before starting work that touches paths another member may declare, or when a peer sends you a plan body.
Skill Claude Code
Score a judgment call with trwassess (calibrated probabilities) instead of settling it in prose. Use at a fork with two or more defensible options; to triage whether a failure is real or a flake / harness artifact, whether a review finding is a blocker or a follow-up, or whether evidence is sufficient to call work…
Skill Claude CodeCodex
Writes and receives Agent Handoff Records: a tiered, sealed JSON record with labelled claims, verbatim constraints and digest-bound pointers, so another session, agent or harness can continue unfinished work after a read-back. Use when asked for a "handoff", to "hand off" or "hand this over", to "write a handoff for…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: