Agent Skills: Project Doc Bootstrap Refresh

Analyze the current repository and create or refresh the project bootstrap docs (`.claude/PROJECT.md`, `workflow/README.md`, `docs/design/README.md`, `docs/glossary/README.md`, `docs/daily/README.md`, `TEST_GUIDE.md`, `docs/analysis/README.md`) to match the project.

UncategorizedID: munlucky/claude-settings/project-md-refresh

Install this agent skill to your local

pnpm dlx add-skill https://github.com/munlucky/moonshot-relay/tree/HEAD/skills/project-md-refresh

Skill Files

Browse the full folder contents for project-md-refresh.

Download Skill

Loading file tree…

skills/project-md-refresh/SKILL.md

Skill Metadata

Name
project-md-refresh
Description
Analyze the current repository and create or refresh the project bootstrap docs (`.claude/PROJECT.md`, `workflow/README.md`, `docs/design/README.md`, `docs/glossary/README.md`, `docs/daily/README.md`, `TEST_GUIDE.md`, `docs/analysis/README.md`) to match the project.

Project Doc Bootstrap Refresh

Goal

Create or update the project bootstrap doc set with accurate, evidence-based project details.

This skill is a bootstrap generator, not a gate by itself. Gate logic belongs in project-contract-gate. This is a maintenance utility, not part of the default implementation chain.

Workflow

  1. Locate the base files.

    • If .claude/PROJECT.md exists, use it as the base; preserve custom rules and update facts.
    • If missing, create .claude/ if needed and copy assets/PROJECT.template.md to .claude/PROJECT.md.
    • For project reference docs, create missing files from these assets:
      • assets/WORKFLOW.README.template.md -> workflow/README.md
      • assets/DESIGN.README.template.md -> docs/design/README.md
      • assets/GLOSSARY.README.template.md -> docs/glossary/README.md
      • assets/DAILY.README.template.md -> docs/daily/README.md
      • assets/TEST_GUIDE.template.md -> TEST_GUIDE.md
      • assets/ANALYSIS.README.template.md -> docs/analysis/README.md
  2. Collect signals from the repository.

    • Read README.md and any project docs.
    • Identify the stack from config files: package.json, pyproject.toml, go.mod, Cargo.toml, build.gradle, pom.xml, Gemfile, requirements.txt, Makefile, or Taskfile.
    • Extract run/build/test/lint commands from scripts or tooling.
    • Inspect the top-level directory structure and key entrypoints.
    • Locate API routes/controllers and data models (search for route, router, controller, handler, model, schema).
    • Find auth configuration (search for auth, jwt, session, oauth).
    • Note environment variable usage (search for ENV, process.env, os.environ, dotenv).
    • Identify shared design systems, token files, component libraries, or UI primitives.
    • Identify canonical domain terminology from product docs, route names, navigation labels, and schema names.
    • Identify ambiguous or conflicting domain terms and choose canonical aliases only when evidence supports the choice.
    • Identify the real test entrypoints, CI checks, smoke/regression commands, and any manual QA flows.
    • Identify workflow expectations such as branch naming, worktree usage, required scripts, release flow, or PR rules from existing docs or scripts.
  3. Update the bootstrap doc set.

    • .claude/PROJECT.md
    • Fill the overview (name, stack, primary language).
    • Summarize core rules and conventions.
    • Record project-local adapter boundaries, including project identity and knowledge/verification contract locations when present.
    • Document directory structure (top-level + key subdirectories).
    • Document API/data patterns, auth, and docs paths.
    • Add concrete commands for dev/build/lint/test/typecheck.
    • workflow/README.md
      • Record the official development flow, role split, standard entry scripts/commands, and document priority.
    • docs/design/README.md
      • Record shared UI/system rules, component or token conventions, and how to handle new design patterns.
    • docs/glossary/README.md
      • Record canonical terms, definitions, aliases to avoid, relationships between terms, and the update rule for new terms.
      • Keep terms domain-facing; skip class/function names unless they are also product language.
      • Add TODOs instead of inventing definitions when the repository uses a term inconsistently.
    • docs/daily/README.md
      • Record daily log structure, required files, minimum required events, and escalation/handoff notes.
    • TEST_GUIDE.md
      • Record test commands, scope strategy, environments, manual QA rules, and what must run before completion.
    • docs/analysis/README.md
      • Record how to structure impact analysis, architecture notes, and investigation writeups.
  4. Output.

    • Save the refreshed bootstrap docs.
    • Provide a short summary and list any gaps/questions.
    • Report whether the minimum contract sections are now ready for orchestration:
      • overview
      • commands
      • testing rules
      • structure/patterns
      • git workflow
      • core rules / boundaries
      • project reference docs

Guardrails

  • Do not invent details; base every statement on files found.
  • If information is missing, add TODOs or ask for confirmation.
  • Keep content concise and project-specific.
  • Preserve user-written project docs; extend or refresh them instead of rewriting from scratch when possible.
  • Do not create ADRs during bootstrap unless a hard-to-reverse, surprising, trade-off decision is already documented by the project.

Project Knowledge Boundary

Bootstrap docs may define or point to .claude/project.identity.yaml, .claude/knowledge.contract.yaml, .claude/verification.contract.yaml, and project prompt files. They must not import raw MemoryGraph/KG/ontology/log/transcript payloads. If knowledge state is unavailable, write TODOs and contract references instead of fabricating facts.