grasdouble/gd-shared-update-docs-project-standard

gd-shared-update-docs-project-standard

Overview

This skill maintains up-to-date documentation in {docs_output_path}/ for a standard single-repo project (not a monorepo). It delegates all analysis and writing to a single subagent that treats the entire repository as one documentation unit — no package splitting, no parallel subagents.

Why a subagent instead of doing it directly? A dedicated subagent gets a fresh, isolated context scoped to the documentation task — no cross-contamination with the orchestration conversation, better output quality.

On Activation

1. Load config via bmad-init skill — Use module bmm:

   Use bmad-init skill:
   - module: bmm
   

   Store all returned vars for use throughout the skill:

   • {user_name} — for greeting
   • {communication_language} — for all communications
   • {document_output_language} — for generated docs
   • {project_name} — project identifier
   • {project_knowledge} → alias as {docs_output_path} — docs output folder (e.g. _bmad-docs)

2. Greet {user_name} in {communication_language}, briefly explain the workflow.

3. Route to stage 00-init.md to begin — load and execute it.

   If config is missing or incomplete, infer values from the project at runtime (read root package.json or other manifest). Only ask the user if inference is impossible.

Stages

| # | Stage | Purpose | Prompt | |---|-------|---------|--------| | 0 | init | Run scan, display result, capture user decision (Update / Skip) | 00-init.md | | 1 | update-doc | Generate or update the single project documentation file | 01-update-doc.md |

Reference

• references/universal-rules.md — Rules that apply to every step

Assets

• assets/subagent-doc.prompt.md — Subagent prompt template: project documentation (with {{variable}} substitution)
• assets/project-doc.template.md — Output template for the project documentation file

Scripts

• scripts/scan-project.py — Deterministic project scanner. Detects if any source files have changed since the last generatedAtCommit stored in {docs_output_path}/project.md. Run from project root via python3 scripts/scan-project.py [--root /path/to/repo]. Base branch auto-detected; override with --base-branch if needed. Outputs JSON to stdout.

Constraints

• Single-session only: All workflow state is held in the LLM's context memory. There is no persistence between sessions. If the conversation resets, the skill must be restarted from 00-init.md.
• No partial resume: Interrupted runs cannot be continued in a new session.

Advanced Usage

Headless Mode (CI/CD)

Set {headless} = true in the skill config vars to skip the HITL confirmation:

Use bmad-init skill:
- module: bmm
- vars: ...,headless:true

Headless auto-decision: 00-init.md — if scan detects changes, auto-proceeds to 01-update-doc.md. If up-to-date, workflow ends.