my-claude

My Claude

Generates documentation about your Claude Code installation and environment.

Prompt contract

Position 1 of $ARGUMENTS is a free-form prompt (RU/EN) — the mode prefixes below are optional and still work standalone. Nobody types keys: resolve mode + query FROM the prompt.

An explicit prefix (ext, external, r , research ) anywhere wins outright, no scoring — see the literal-syntax table below.
Else score modes by distinct whole-word keyword hits (table below). Highest unique score wins; tie -> AskUserQuestion; all zero -> internal (documented default).
Empty arguments -> internal. Read-only inventory + write to ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/ — asks nothing.
Outcome-changing ambiguity (e.g. an existing INDEX entry for the same mode) -> ONE AskUserQuestion.
Prose that is not a mode/query is still input: extract the topic/query from it.

Then print this block ONCE, before the first action:

PLAN — brewdoc:my-claude
INPUT:  <arguments verbatim, or "(empty)">
MODE:   <resolved> — <explicit prefix | matched keyword: X | default>
SCOPE:  <sources to analyze | research query>
DO:     <2-5 imperative bullets>
RESULT: <doc path(s) written under ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/>

Labels are literal; authored work-artifact values are English; preserve INPUT verbatim.

Mode Detection

Mode	EN keywords	RU keywords	Mutates?
`internal` (DEFAULT)	(empty), internal, my setup, my installation, document my claude	internal, my setup, document my claude	yes
`external`	ext, external, architecture, hook/context schema	external, architecture, context diagram	yes
`research`	r, research, look up, find out about	research, investigate, find information	yes

Literal prefix syntax (still accepted, wins outright over scoring):

`$ARGUMENTS` value	Mode	Sub-mode
empty	INTERNAL	—
`ext` or `external` (alone)	EXTERNAL	default
`ext context` or `external context`	EXTERNAL	context-schema
starts with `r` or `research`	RESEARCH	query = rest of args

After detection, load the appropriate reference file:

INTERNAL: references/internal-mode.md
EXTERNAL: references/external-mode.md
RESEARCH: references/research-mode.md

Reference files are read raw — no substitution runs on them. Where one prints ${CLAUDE_PROJECT_DIR}, use the project root this skill's own markdown already resolved.

vs built-in `/team-onboarding`

Built-in /team-onboarding (CC 2.1.101+) is enough for a quick teammate handoff doc from local config. Use this skill instead when the job needs web research, EXTERNAL architecture synthesis, RESEARCH mode, or the persistent ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/INDEX.jsonl with citations.

Delegation

One subagent = one source group, ~<=5 files, ~<=10 steps. Split larger work into N tasks, spawned in ONE message; an hour-long agent drifts beyond observation/correction.

Every spawn prompt MUST carry:

Field	Content
GOAL	the overall task and why it exists — the point beyond the file edit
ROLE	what this agent owns; what it must NOT touch
SCOPE	exact paths/commands in bounds + explicit out-of-bounds
CONTEXT	what is already done, by whom, what runs in parallel — trimmed to what THIS agent needs
CONSUMER	who or what uses the result next, and the shape it must fit
DONE	acceptance criteria + the exact report shape you want back

A bare one-line task is never enough. Applies to every Explore / general-purpose spawn in all three modes.

Output Directory

All generated docs go to ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/. Use the token in EVERY output and index path, never a bare relative one: skill commands run in the session shell's cwd and that cwd moves with cd, so a bare path invoked from a nested package builds a second output tree and a second INDEX. Create if not exists: mkdir -p "${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude"

Spell it BARE — ${CLAUDE_PROJECT_DIR}. The token is a text substitution into this skill's markdown and its allowed-tools (CC 2.1.196+), not an env var; a modifier form like ${CLAUDE_PROJECT_DIR:-...} is not matched, reaches the shell verbatim, and always falls through to its fallback.

This is the only supported target — there is no ~/.claude or plugin-data fallback. ~/.claude/** is an ASK-classified path: it prompts in default/acceptEdits, is auto-approved under bypassPermissions, and a headless run without bypass fails on it.

Provenance frontmatter (every generated doc, all three modes)

Every .md this skill writes opens with this block, before the # heading. doc_type is BARE; the other three are QUOTED. Resolve the values — never hardcode them.

EXECUTE using Bash tool before writing the document:

PJ="${CLAUDE_SKILL_DIR}/../../.claude-plugin/plugin.json"
PV=$(node -e "process.stdout.write(JSON.parse(require('fs').readFileSync(process.argv[1],'utf8')).version||'')" "$PJ" 2>/dev/null || true)
[ -n "$PV" ] || { echo "❌ cannot read version from $PJ — reinstall brewdoc"; exit 1; }
echo "PLUGIN_VERSION=$PV"; echo "LAST_UPDATED=$(date +%F)"

---
doc_type: user
version: "{PLUGIN_VERSION}"
generated_by: "brewdoc:my-claude"
last_updated: "{LAST_UPDATED}"
---

Re-generating an existing doc REFRESHES all three quoted values and leaves a hand-edited doc_type (llm / skip) as the user set it.

INDEX Tracking

Append entry to ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/INDEX.jsonl. The path field stays repo-relative — the token anchors the file you open, not the value you record:

{"ts":"2026-02-28T10:00:00","mode":"internal","path":".claude/brewdoc/my-claude/20260228_my-claude-internal.md","title":"Internal Claude Setup Overview","version":"1.0"}

Legacy read-only merge — if ~/.claude/brewdoc/INDEX.jsonl exists AND the new project INDEX is empty, read the legacy file once, merge its entries into ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/INDEX.jsonl, and print: ℹ️ Migrated {N} entries from legacy ~/.claude/brewdoc/INDEX.jsonl (read-only; legacy file untouched). NEVER write back to the legacy path.

If an existing entry for the same mode exists: use AskUserQuestion — header: "INDEX", question: "Entry for this mode already exists (v{VERSION}). Update it?", options: "Yes, update (bump version)" / "No, create new entry".

INTERNAL Mode

Goal: Document your local Claude Code setup — CLAUDE.md files, rules, agents, skills, memories.

Sources to analyze:

~/.claude/CLAUDE.md — global instructions
~/.claude/rules/*.md — global rules
~/.claude/agents/*.md — global agents
~/.claude/skills/ — global skills
Project CLAUDE.md (current working directory)
.claude/rules/*.md — project rules
~/.claude/projects/**/memory/MEMORY.md — memory files

Process:

Spawn 3 parallel Explore agents, one per source group: (1) global ~/.claude config, (2) project .claude config, (3) memory files. Brief each in full shape, e.g. group (1):

Agent(subagent_type="Explore", prompt="
GOAL: producing a document describing this user's whole Claude Code installation;
  you cover the global ~/.claude layer only, another agent covers project + memory.
ROLE: read-only inventory of global config. Do NOT edit any file, do NOT read project files.
SCOPE: in — ~/.claude/CLAUDE.md, ~/.claude/rules/*.md, ~/.claude/agents/*.md, ~/.claude/skills/.
  Out — project CLAUDE.md, .claude/**, memory files, source code.
CONTEXT: nothing has been read yet — you are the first pass for this group. Two sibling
  agents scan project `.claude` config and memory files in parallel right now; the output
  dir `${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/` and the INDEX entry are already
  resolved by the skill.
CONSUMER: the aggregation step merges the three inventories into one INTERNAL-mode
  document, then an `Explore` agent re-checks that every path you name exists — absolute
  paths only, and flag a missing file instead of dropping its row.
DONE: per-section inventory (instructions summary, rules N + one line each, agents N + purpose,
  skills N + purpose) with absolute paths for every item. Flag missing files explicitly.
")

Aggregate findings into structured document
Write to ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/YYYYMMDD_my-claude-internal.md
Spawn an independent Explore agent to validate facts (file paths exist, content accurate) — read-only check, no edits
Apply the validation fixes if any
Add INDEX entry

Output document structure:

# Claude Code Internal Setup — {date}

## Global Configuration
### Instructions (CLAUDE.md)
### Rules ({N} rules)
### Agents ({N} agents)
### Skills ({N} skills)

## Project Configuration
### Project Instructions
### Project Rules

## Memory
### Active Memories ({N} entries)

## Summary
| Component | Count | Location |
|-----------|-------|----------|

EXTERNAL Mode

Goal: Document Claude Code's hook/context/agent architecture from official sources + local analysis.

Sub-mode default:

Analyze local hook files for event model patterns
WebSearch for recent Claude Code releases and CHANGELOG
Spawn general-purpose agents for: official docs (code.claude.com), GitHub releases, community forums
Generate ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/YYYYMMDD_my-claude-external.md

Sub-mode context-schema:

Focus specifically on context injection schema (additionalContext, updatedInput, etc.)
Output: ${CLAUDE_PROJECT_DIR}/.claude/brewdoc/my-claude/external/YYYYMMDD_context-schema.md

RESEARCH Mode

Goal: Research a specific query about Claude Code using multiple sources.

Query: everything after r or research in $ARGUMENTS