oat-project-document

name	oat-project-document
version	1.6.2
description	Use when the user requests or confirms documenting an active OAT project — e.g. "document the project", "update the docs", "run oat-project-document", or confirms a previously offered documentation run. Do NOT auto-invoke when implementation completes. Analyzes project artifacts, presents a documentation delta plan, and applies approved changes.
argument-hint	[project-path] [--auto]
disable-model-invocation	false
user-invocable	true
allowed-tools	Read, Write, Edit, Bash(git:*), Bash(oat tools:*), Glob, Grep, AskUserQuestion, Skill
metadata	{"internal":true}

Project Documentation Sync

Read project artifacts and implementation code to identify documentation surfaces that need updating, present a delta plan for approval, and apply changes — all in a single invocation.

Prerequisites

Required:

• Active OAT project (or explicit project path) with at least plan.md or implementation.md
• Project should have completed some implementation work (artifacts describe what was built)

Mode Assertion

OAT MODE: Project Document

Purpose: Analyze what a project built, identify documentation gaps, and apply approved documentation updates.

When OAT_AUTONOMOUS=1, read references/docs/autonomy-contract.md and execute the existing --auto path for this run. Keep OAT_NON_INTERACTIVE=1 session-scoped and never persist either environment signal. This branch is inert when autonomy is inactive.

Progress Indicators (User-Facing)

• Print a phase banner once at start:

  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OAT ▸ PROJECT DOCUMENT ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

• For each step, announce a compact header:

  • OAT ▸ DOCUMENT — Step N: {step_name}

• Before multi-step work:

  • [1/N] {action}…

• Keep it concise; don't print a line for every shell command.

BLOCKED Activities:

• No modifying implementation source code
• No modifying project phase state (except oat_docs_updated in state.md)
• No deleting or restructuring existing documentation without user approval
• No creating documentation content that contradicts source code

ALLOWED Activities:

• Reading all project artifacts (discovery.md, spec.md, design.md, plan.md, implementation.md)
• Reading source code referenced in artifacts
• Scanning all documentation and instruction surfaces
• Writing/editing documentation files (after approval or with --auto)
• Creating new documentation files and directories
• Editing docs tooling config (e.g., mkdocs.yml nav) for new files
• Updating oat_docs_updated in state.md

Self-Correction Protocol: If you catch yourself:

• Modifying source code → STOP (this skill only writes documentation)
• Inventing documentation without evidence → STOP (every recommendation needs artifact/code evidence)
• Changing project phase or task state → STOP (only oat_docs_updated is allowed)

Recovery:

1. Acknowledge the deviation
2. Return to current step
3. Ask user for guidance

Artifact Hygiene

Artifact hygiene contract: Before finishing or committing, format every file you created or edited. Use the concrete write/fix formatting command supplied by the governing plan, task, or brief. If none is usable, discover the repository's documented write/fix command from applicable AGENTS.md/CLAUDE.md instructions and relevant package manifests; do not infer or hardcode a formatter. Prefer a file-scoped invocation when supported, and avoid rewriting unrelated files. If no command is discoverable, warn once with no format command discovered in repo instructions; skipping, then continue.

After formatting, run only repository checks relevant to the files changed; writing documentation does not imply unrelated full test suites.

Argument Parsing

Parse $ARGUMENTS for:

1. --auto flag: If present, skip user approval and apply all recommendations directly.

   AUTO_MODE=false
   if echo "$ARGUMENTS" | grep -q '\-\-auto'; then
     AUTO_MODE=true
   fi
   

   If OAT_AUTONOMOUS=1, set AUTO_MODE=true after normal argument parsing even when --auto was not passed. Record gate DOCUMENT-02 and the applied recommendation set in the documentation outcome. This is the existing --auto behavior, not a new approval mode.

2. project-path: Any non-flag argument is treated as an explicit project path.

   PROJECT_ARG=$(echo "$ARGUMENTS" | sed 's/--auto//g' | xargs)
   

Process

Step 0: Resolve Project Context

OAT stores active project context in .oat/config.local.json (activeProject, local-only).

PROJECT_PATH=$(oat config get activeProject 2>/dev/null || true)
PROJECTS_ROOT="${OAT_PROJECTS_ROOT:-$(oat config get projects.root 2>/dev/null || echo ".oat/projects/shared")}"
PROJECTS_ROOT="${PROJECTS_ROOT%/}"

Resolution order:

1. If $PROJECT_ARG is provided, use it as $PROJECT_PATH
2. Else if activeProject is set in config, use that
3. If neither available, use AskUserQuestion to ask: "No active project found. Please provide the path to the project directory (e.g., .oat/projects/shared/my-project)."

Validation:

• Verify $PROJECT_PATH/state.md exists
• Verify at least one of plan.md or implementation.md exists (need to know what was built)
• If validation fails, report the error and STOP

Derive {project-name} as the directory name: basename "$PROJECT_PATH".

Documentation config resolution:

Read documentation config from .oat/config.json:

DOCS_ROOT=$(oat config get documentation.root 2>/dev/null || true)
DOCS_TOOLING=$(oat config get documentation.tooling 2>/dev/null || true)
DOCS_CONFIG=$(oat config get documentation.config 2>/dev/null || true)

If $DOCS_ROOT is empty, attempt auto-detection:

1. Look for mkdocs.yml — if found, set DOCS_TOOLING=mkdocs and infer DOCS_ROOT from the docs_dir field (or default to docs/ relative to the mkdocs.yml location), set DOCS_CONFIG to the mkdocs.yml path
2. Look for docusaurus.config.js or docusaurus.config.ts — if found, set DOCS_TOOLING=docusaurus, infer DOCS_ROOT from config
3. Look for conf.py (Sphinx) — if found, set DOCS_TOOLING=sphinx
4. Look for a top-level docs/ directory — if found, set DOCS_ROOT=docs
5. If nothing found, DOCS_ROOT remains empty (docs directory scanning will be skipped)

For auto-detection, use Glob to scan from repo root:

Glob: **/mkdocs.yml (exclude node_modules, .oat, dist)
Glob: **/docusaurus.config.{js,ts}

Store resolved values for use in later steps. Do not write auto-detected values to config.

Step 1: Check for PJM Infrastructure

Check whether the project-management tool pack is effectively available:

PJM_INSTALLED=$(oat tools has project-management 2>/dev/null || echo "false")

If PJM_INSTALLED is true:

• Invoke oat-pjm-update-repo-reference automatically before proceeding.
• Do not ask whether to run the repo reference update during project-document.
• If invocation succeeds: Log that repo reference docs were refreshed and continue to Step 2. The reference surfaces will already be current when the documentation scan reads them in Step 4a.4.
• If invocation fails: Warn the user that the repo reference update failed, but continue with the documentation sync — a PJM failure should not block documentation updates.

If PJM_INSTALLED is not true: Skip silently and proceed to Step 2.

Step 2: Read Project Artifacts

Read all available project artifacts to build an understanding of what was built.

Read in order:

1. $PROJECT_PATH/discovery.md — initial requirements, key decisions, constraints
2. $PROJECT_PATH/spec.md (if exists) — formal requirements, acceptance criteria
3. $PROJECT_PATH/design.md (if exists) — architecture, components, data models, integration points
4. $PROJECT_PATH/plan.md — phases, tasks, file lists, commit messages
5. $PROJECT_PATH/implementation.md — execution log, outcomes, files changed, decisions

Synthesize a "what was built" model:

From the artifacts, extract and organize:

• Features and capabilities added — what can users/developers now do that they couldn't before?
• Architectural decisions — new patterns, component boundaries, data flow changes
• New frameworks, tooling, or libraries — dependencies added, build tool changes, test frameworks
• New CLI commands or config schema changes — new user-facing commands, config keys, options
• New directories or structural changes — new packages, apps, or significant directory reorganization
• API changes — new endpoints, modified interfaces, changed contracts

Note source file references:

While reading artifacts, collect all source file paths mentioned in:

• Plan task **Files:** sections (Create/Modify entries)
• Implementation.md **Files changed:** entries
• Design component interfaces and data models

These will be verified against actual code in Step 3.

Handle missing artifacts gracefully:

• Quick-mode projects may lack spec.md and design.md — extract what's available
• If only plan.md exists (no implementation.md), the project may not have started implementation yet — still proceed, but note that documentation recommendations will be based on planned work rather than verified implementation

Step 3: Verify Against Code and Build a Capability Inventory

Read source files referenced in artifacts to confirm what actually shipped, then do a targeted capability-discovery pass so the skill can catch newly introduced documentation surfaces rather than only updating already-documented ones.

For each referenced source file:

1. Check if the file exists (it may have been renamed or deleted since the plan was written)
2. Read the file to verify artifact claims:
   • Do described APIs/commands/config schemas actually exist?
   • Were architectural patterns from design.md followed?
   • Are there implementation details not captured in artifacts?

Augment the model:

• Add code-verified details that artifacts didn't capture
• Note any discrepancies between artifacts and code (informational — include in delta plan as context, not as blocking issues)
• Organize the verified implementation into capability areas (for example: new app surface, CI/CD pipeline, release automation flow, deployment target, integration surface, CLI/config workflow)

Targeted capability discovery pass:

After verifying artifact-referenced files, inspect the strongest adjacent signals for newly shipped capability surfaces even when the exact docs target does not already exist.

Prioritize evidence from:

• new apps/packages/directories called out by the artifacts
• workflow/config files tied to shipped behavior (.github/workflows, release configs, deploy configs, mobile build/release files, etc.)
• entrypoints, route registration files, service modules, and schemas that define user- or operator-facing behavior
• package manifests and scripts that expose new setup, release, or operational workflows

For each significant capability area, capture:

• capability name
• concrete repo evidence proving it shipped
• likely audience (developer, operator, integrator, end user)
• whether the capability represents a new docs surface versus an addition to an existing surface

Scope control:

• Start with files directly referenced in artifacts, then inspect only the highest-signal adjacent files needed to understand the shipped capability areas
• Do not scan the entire codebase blindly; stay anchored to the implementation areas surfaced by the project artifacts
• If artifacts reference many files (>20), prioritize: new files first, then modified files with the most changes, then only the adjacent files needed to confirm docs impact
• Read file contents, not just check existence — the skill needs to understand what the code does to make good documentation recommendations

Step 4: Discover Documentation Surfaces

Scan the repository for all documentation and instruction surfaces.

4a. Documentation surfaces (primary — thorough analysis):

1. Docs directory (if $DOCS_ROOT is set):

   • Read the docs tooling config (e.g., $DOCS_CONFIG) to understand nav structure
   • List all files in $DOCS_ROOT recursively
   • Read existing docs files that could be affected by the project
   • Identify the parent section or directory where each uncovered capability area would naturally live
   • Note when no existing page or directory is a good fit — this is a strong signal for a CREATE recommendation, not a reason to force the content into an unrelated existing page

2. Root README.md:

   • Always check — read current content

3. Subdirectory README.md files:

   • Glob **/README.md (exclude node_modules, .oat, dist, .worktrees)
   • Read existing ones that are in directories affected by the project
   • Note directories for new apps/packages that lack a README

4. Reference files:

   • Check the active PJM layer under .oat/repo/pjm/ and the durable layer under .oat/repo/reference/
   • Read (whichever exist): pjm/current-state.md, pjm/roadmap.md, pjm/backlog/index.md, pjm/backlog/completed.md, relevant pjm/backlog/items/*.md files, and reference/decisions/index.md plus relevant reference/decisions/*.md records
   • Legacy repos may still have reference/current-state.md, reference/roadmap.md, reference/backlog/, or a reference/decision-record.md monolith until they run oat pjm migrate; read those only as a fallback when the pjm/ and reference/decisions/ equivalents are absent

4b. Instruction surfaces (secondary — strong signals only):

1. Root AGENTS.md / CLAUDE.md:

   • Always check — read current content

2. Subdirectory AGENTS.md files:

   • Glob **/AGENTS.md (exclude node_modules, .oat, dist, .worktrees)
   • Read ones in directories affected by the project

3. Provider rules files:

   • Check .oat/sync/config.json for enabled providers
   • For each enabled provider, check its rules directory:
     • Claude: .claude/rules/
     • Cursor: .cursor/rules/
     • Copilot: .github/copilot-instructions.md
     • Gemini: .gemini/rules/
   • Read existing rules files that may need updating

Store surface inventory for use in Step 5. For each surface, record:

• File path (existing or potential)
• Surface type (docs | readme | reference | agents | provider-rules)
• Current content summary (for existing files)
• Whether it's in a directory affected by the project

Step 5: Assess Documentation Delta

Compare "what was built" (from Steps 2-3) against "what's documented" (from Step 4) to produce recommendations.

5a. Capability coverage assessment (required):

Before recommending file-level edits, evaluate coverage for each significant capability area in the "what was built" model.

For each capability area, classify the documentation state as:

• adequately covered — existing docs already explain the shipped behavior accurately
• thin coverage — the area is mentioned, but important setup, workflow, or usage details are missing
• no coverage — no existing docs surface meaningfully covers the capability

For each capability with thin coverage or no coverage, determine the best docs home:

• Expand an existing page when a clear, discoverable page already owns the topic
• Create a new page when the topic is distinct enough to deserve its own entrypoint
• Create a new docs directory (with index.md entrypoint, where that is the local convention) when the shipped work introduces a new top-level capability area with multiple likely subtopics

Bias rules:

• If a docs app exists and the shipped work represents a new major capability area, do not default to stuffing the information into README.md, roadmap.md, or current-state.md
• Examples that often warrant CREATE recommendations: new mobile apps, CI/CD and release automation, new deployment targets, new integration surfaces, or new operator workflows
• If no existing docs section is a natural fit, prefer recommending a new page or directory rather than forcing an UPDATE on a loosely related page

Each coverage-gap finding should capture:

• capability area
• current docs state (thin coverage or no coverage)
• evidence proving the capability shipped
• likely audience
• suggested docs location (existing page, new page, or new directory)
• why that location is the right home

5b. Documentation surface assessment:

For each documentation surface relevant to the project, determine one of:

• UPDATE: Existing doc needs content changes. Specify:

  • What content to add/modify
  • Where in the file it should go
  • Why (evidence from artifacts/code)

• CREATE: No existing doc covers this area. Specify:

  • Proposed file path
  • Proposed content outline (section headings, key points)
  • Why this warrants a new file (evidence)
  • If inside a docs app, note any parent index/nav follow-on work that will be needed

• SPLIT: Existing doc would become too large with additions. Specify:

  • Current file path and approximate size
  • Proposed split structure (which content moves where)
  • Why splitting is recommended (file would exceed ~300 lines, or logical separation warrants it)

• No change: Surface is already accurate — skip from delta plan.

When deciding between UPDATE and CREATE, prefer CREATE if:

• the capability currently has no coverage
• the docs app has no discoverable home for that topic
• the proposed addition would otherwise bury a new major workflow inside an unrelated page

5c. Instruction surface assessment (strong signals only):

Only recommend instruction changes when there is a clear trigger:

Signal	Example	Recommendation
New test framework	vitest added to devDependencies	Create test rules for enabled providers
New styling/component library	tailwind, storybook added	Create styling rules for enabled providers
New build tooling	different bundler, new build step	Update AGENTS.md development commands
New directory with complex patterns	new package with unique conventions	Create subdirectory AGENTS.md
New dev commands	new CLI commands, scripts	Update root AGENTS.md

If no strong signal is present for an instruction surface, skip it.

5d. Per recommendation, capture:

- Target: {file path — existing or proposed}
- Action: {UPDATE | CREATE | SPLIT}
- Summary: {1-2 sentences on what changes and why}
- Evidence: {artifact/code reference — e.g., "spec.md §3", "plan.md p02-t03", "implementation.md p01-t01 outcome", "workflow file X proves release automation shipped"}
- Audience: {developer | operator | integrator | end user}
- Content guidance: {specific content to add or outline for new files}
- Coverage state: {adequately covered | thin coverage | no coverage}
- Parent docs impact: {parent index/nav updates needed, or "none"}

Step 6: Present Delta Plan

Format and present the recommendations for user approval.

6a. Format output:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 OAT ▸ PROJECT DOCUMENT — Delta Plan
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## Documentation Updates ({N} recommendations)

1. {ACTION}  {target file path}
   {Summary of what changes and why}
   Evidence: {artifact/code reference}

2. {ACTION}  {target file path}
   {Summary}
   Evidence: {reference}

## Instruction Updates ({N} recommendations)

3. {ACTION}  {target file path}
   {Summary}
   Evidence: {reference}

6b. Handle edge cases:

• If $AUTO_MODE is true: skip to Step 7 (apply all recommendations)
• If no recommendations found: report "No documentation updates identified for this project.", set oat_docs_updated: complete in state.md, and exit
• If only instruction recommendations (no docs): still present, but note the documentation-first priority

Under OAT_AUTONOMOUS=1, the auto path applies all non-destructive recommendations without presenting Step 6c. A recommendation that deletes documentation or broadly restructures existing documentation beyond the approved project plan remains gate DOCUMENT-03: stop before applying it, record the proposed change and evidence, and leave the project resumable.

6c. Interactive approval:

Approve recommendations?
  [Y]es — apply all
  [I]ndividual — approve/reject each one
  [S]kip — skip documentation updates

• Yes: mark all recommendations as approved
• Individual: present each recommendation one at a time with approve/reject
• Skip: set oat_docs_updated: skipped and oat_project_state_updated: "{ISO 8601 UTC timestamp}" in $PROJECT_PATH/state.md frontmatter, commit the state change, and exit without applying documentation changes

Track which recommendations were approved for Step 7.

Step 7: Apply Approved Changes

Execute the approved documentation updates.

For each approved recommendation:

1. UPDATE:

   • Read the target file
   • Edit the file to add/modify content as specified in the recommendation
   • Preserve existing content structure — insert new sections or update existing ones

2. CREATE:

   • Create parent directories if needed (mkdir -p)
   • Write the new file with the content outlined in the recommendation
   • For docs directory files: follow the existing docs conventions for entrypoints and local navigation (for example, index.md entrypoints and ## Contents sections where that is the local pattern)
   • If the recommendation created a new docs directory, add the required entrypoint file for that directory as part of the same change

3. SPLIT:

   • Create the new file with the content being moved
   • Edit the original file to remove the moved content
   • Add a cross-reference in the original file pointing to the new location
   • If the original had sections that logically separate, use section headings as split boundaries

Nav structure updates:

If $DOCS_CONFIG exists and new files were created in the docs directory:

• Read the tooling config (e.g., mkdocs.yml)
• Add new entries to the nav structure in the appropriate location
• Preserve existing nav order

Error handling:

• Track a $ALL_SUCCEEDED flag (default: true). If any file write fails, set $ALL_SUCCEEDED to false, log the error, and continue with remaining recommendations
• At the end, report any failures with the specific files that could not be written

Step 8: Commit and Update State

8a. Stage and commit documentation changes:

git add {list of changed/created documentation files}
git diff --cached --quiet || git commit -m "docs({project-name}): update documentation from project artifacts"

Only stage files that were actually changed or created in Step 7. Do not use git add -A.

8b. Update project state:

Update $PROJECT_PATH/state.md frontmatter based on apply outcome:

• If $ALL_SUCCEEDED is true: set oat_docs_updated: complete and oat_project_state_updated: "{ISO 8601 UTC timestamp}"
• If $ALL_SUCCEEDED is false: do not set oat_docs_updated: complete — leave the field as null so the skill can be re-run. Still set oat_project_state_updated: "{ISO 8601 UTC timestamp}". Surface the failures clearly in the summary report (Step 8d) so the user knows which updates failed and why.

git add "$PROJECT_PATH/state.md"
git diff --cached --quiet || git commit -m "chore({project-name}): mark docs updated"

8c. Handle edge cases:

• If user explicitly skipped (chose [S]kip in Step 6): oat_docs_updated was already set to skipped in Step 6. No further state update needed here.
• If no recommendations were found: set oat_docs_updated: complete (nothing to do is still "done").
• If --auto mode: apply all, commit, set state — no user interaction.

8d. Report summary:

If $ALL_SUCCEEDED is true:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 OAT ▸ PROJECT DOCUMENT — Complete
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Documentation sync complete for {project-name}.

  Updated: {N} files
  Created: {N} files
  Split:   {N} files

Commit: {sha}
State:  oat_docs_updated = complete

Next steps:
  - oat-project-complete → close out the project
  - Review changes before pushing

If $ALL_SUCCEEDED is false:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 OAT ▸ PROJECT DOCUMENT — Partial Failure
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Documentation sync partially failed for {project-name}.

  Succeeded: {N} files
  Failed:    {N} files
    - {path}: {error reason}

Commit: {sha} (successful changes only)
State:  oat_docs_updated NOT set (re-run to retry)

Next steps:
  - Investigate and fix the failed writes
  - Re-run oat-project-document to complete the sync

Success Criteria

• All documentation surfaces relevant to the project are scanned
• Significant capability areas from the "what was built" model are classified as adequately covered, thin coverage, or no coverage before file-level recommendations are chosen
• Recommendations are evidence-based (every recommendation cites artifact/code sources)
• CREATE actions are recommended when no existing docs surface is a natural fit, including new docs pages or directories when the shipped work introduces a new capability area
• Interactive approval flow works correctly (all/individual/skip)
• --auto mode applies all changes without user interaction
• New files and splits are handled correctly
• Docs tooling nav is updated when applicable
• oat_docs_updated state is set correctly in all paths
• Skill works on both active and completed/archived projects
• Skill does not modify source code or project phase state (except oat_docs_updated)