Documentation Maintenance

When to Apply

Activate this skill when:

• •The doc-sync-hint PostToolUse hook flags related skills or docs after a code change
• •You are directly creating or editing files in .claude/skills/, .claude/agents/, or docs/
• •The user asks you to update documentation, skills, or agents
• •You notice that existing skill or agent guidance no longer matches actual code patterns
• •A new domain area needs a dedicated agent, or an existing agent's domain knowledge is outdated

File-to-Domain Mapping

This table shows which skills, docs, and agents relate to each code area. The hook script uses the same mapping for skills and docs.

When to Update Skills

Update a skill's SKILL.md when:

• •A new code pattern or convention is established that the skill should teach
• •An existing convention documented in the skill has changed
• •A new pitfall or gotcha is discovered that future work should avoid
• •New model relationships, enum values, or API endpoints are added to the domain
• •Code snippets in the skill no longer match actual implementations

When NOT to Update Skills

Do not update skills for:

• •Formatting-only changes (Pint ran, whitespace adjustments)
• •Bug fixes that don't change conventions or patterns
• •Adding a single new test that follows existing patterns
• •Trivial additions that match documented patterns exactly (e.g., adding a new tag follows the existing tag pattern)

When to Update Agents

Update an agent's .md file in .claude/agents/ when:

• •The agent's domain knowledge section references patterns or models that have changed
• •New code conventions are established in the agent's domain (e.g., new model relationships, new API patterns)
• •The agent's <example> blocks in the description no longer reflect realistic usage
• •A new tool, service, or integration is added to the agent's domain
• •The agent's memory directory needs seeding with initial knowledge

Do not update agents for:

• •Changes outside the agent's domain (the agent won't be invoked for those areas)
• •Formatting-only or Pint changes
• •Bug fixes that don't change domain patterns

How to Update Skills

SKILL.md Structure

Every skill file follows this structure:

---
name: skill-name
description: >-
  One-paragraph description. Used by Claude to decide when to activate.
---

# Skill Title

## When to Apply
- Bullet list of activation triggers

## Domain sections...
- Code conventions, patterns, relationships
- Use <code-snippet> blocks for examples

## Common Pitfalls
- Things to watch out for

Updating an Existing Skill

• •Make targeted edits — update only the specific section that changed; do not rewrite the entire file
• •Keep the description accurate — it controls when Claude activates the skill; if the domain scope changes, update the description
• •Check hook mappings — if the skill now covers additional code paths, update doc-sync-hint.sh and the mapping table above
• •Use <code-snippet> blocks for code examples — they render correctly in boost guidelines
• •Match sibling skill style — check other skills in .claude/skills/ for formatting conventions

How to Update Docs

What belongs in docs vs skills vs agents

Updating an Existing Doc

• •Update tables and diagrams only when the actual architecture changes (new models, new relationships, changed flows)
• •Keep docs factual and descriptive — they describe what the system is, not how to code it
• •Skills describe how to work with the system — patterns, conventions, pitfalls
• •If a new model or relationship is added, update the relevant schema/architecture doc

How to Update Agents

Agent File Structure

Agent files live in .claude/agents/ and use YAML frontmatter:

---
name: agent-name
description: "When to invoke + <example> blocks showing user/assistant/commentary"
model: sonnet
color: orange
memory: project
---

Agent persona and domain knowledge body...

Updating an Existing Agent

• •Update domain knowledge — when models, relationships, or patterns change in the agent's domain, update the body sections
• •Update <example> blocks — if the agent's invocation patterns change, update examples in the description field
• •Update memory — seed the agent's MEMORY.md at .claude/agent-memory/{agent-name}/MEMORY.md with initial knowledge if it's empty
• •Keep the description scope accurate — the description controls when the Task tool selects this agent
• •Make targeted edits — update specific sections, don't rewrite the entire file

Adding a New Skill

1. •Create .claude/skills/{skill-name}/SKILL.md with YAML frontmatter (name + description)
2. •Add the skill name to the skills array in boost.json
3. •Add path mappings to .claude/hooks/doc-sync-hint.sh if the skill covers specific code paths
4. •Update the mapping table in this file

Adding a New Doc

Root-level docs:

1. •Create docs/{topic}.md (architecture narrative) or add to docs/generated/ (auto-generated references)
2. •Update docs/README.md index
3. •Add path mappings to .claude/hooks/doc-sync-hint.sh for code paths related to the new doc
4. •Update the mapping table in this file

Feature docs:

1. •Create docs/features/{feature-name}/README.md using the /feature skill
2. •Add ## Key References section linking to relevant docs/generated/*.md#feature-name anchors
3. •Update docs/features/index.md

Auto-generated references (docs/generated/):

• •Run php artisan docs:generate to regenerate schema.md, routes.md, jobs.md
• •These are committed but never edited manually — update the command at app/Console/Commands/GenerateReferenceDocs.php instead

Adding a New Agent

1. •Create .claude/agents/{agent-name}.md with YAML frontmatter (name, description, model, color, memory)
2. •Create the agent memory directory: .claude/agent-memory/{agent-name}/MEMORY.md (can be empty initially)
3. •Write the agent body with persona, domain knowledge, implementation patterns, and critical conventions
4. •Include <example> blocks in the description field showing when to invoke the agent (user/assistant/commentary format)
5. •Update the Agents column in the mapping table in this file
6. •No boost.json or hook changes needed — agents are auto-discovered from .claude/agents/

Common Pitfalls

• •Rewriting whole files: Use targeted Edit operations, not Write, when updating existing skills, docs, or agents
• •Wrong file for content: Architecture decisions go in docs/, coding guidance goes in skills, deep domain knowledge goes in agents, project config goes in CLAUDE.md
• •Forgetting hook updates: When adding a new skill or doc, also update the path mappings in doc-sync-hint.sh and the table above
• •Feedback loops: The hook script skips files in .claude/skills/, .claude/agents/, docs/, and CLAUDE.md to prevent infinite hint loops — don't remove those guards
• •Forgetting agent memory directories: When creating a new agent, always create .claude/agent-memory/{agent-name}/MEMORY.md — the agent won't persist learnings without it
• •Mismatched agent description vs domain: The description field determines when the agent is invoked; if it doesn't mention a code area, the agent won't be selected for that work
• •Stale agent examples: When domain patterns change significantly, update the <example> blocks in the agent's description — outdated examples lead to incorrect invocation