/init-deep

Generate hierarchical AGENTS.md files. Root + complexity-scored subdirectories.

Usage

/init-deep                      # Update mode: modify existing + create new where warranted
/init-deep --create-new         # Read existing → remove all → regenerate from scratch
/init-deep --max-depth=2        # Limit directory depth (default: 3)

Workflow (High-Level)

1. Discovery + Analysis (concurrent)
   • Fire background explore agents immediately
   • Main session: bash structure + LSP/ast-grep code map + read existing AGENTS.md
2. Score & Decide - Determine AGENTS.md locations from merged findings
3. Generate - Root first, then subdirs in parallel
4. Review - Deduplicate, trim, validate

Phase 1: Discovery + Analysis (Concurrent)

Mark "discovery" as in_progress.

Fire Background Explore Agents IMMEDIATELY

Don't wait-these run async while main session works. Equip every agent with ast-grep: any task touching structure, entry points, dependencies, or hotspots MUST run the ast-grep skill (sg with $VAR/$$$) or the ast_grep MCP (search/scan), plus lsp_symbols when present, and ground its claims in that data instead of guessing from conventions. WHY: no symbol graph exists, so code shape is matched, not queried. Richer structural context per agent = a more accurate project map.

// Fire all at once, collect results later
task(subagent_type="explore", load_skills=[], description="Explore project structure", run_in_background=true, prompt="Project structure: map real layout via ast-grep structural search (sg/ast_grep MCP) + rg --files → REPORT deviations from standard patterns")
task(subagent_type="explore", load_skills=[], description="Find entry points", run_in_background=true, prompt="Entry points: FIND main files, trace reach via lsp_find_references + ast-grep call-shape search → REPORT non-standard organization")
task(subagent_type="explore", load_skills=[], description="Find conventions", run_in_background=true, prompt="Conventions: FIND config files (.eslintrc, pyproject.toml, .editorconfig) → REPORT project-specific rules")
task(subagent_type="explore", load_skills=[], description="Find anti-patterns", run_in_background=true, prompt="Anti-patterns: FIND 'DO NOT', 'NEVER', 'ALWAYS', 'DEPRECATED' comments → LIST forbidden patterns")
task(subagent_type="explore", load_skills=[], description="Explore build/CI", run_in_background=true, prompt="Build/CI: FIND .github/workflows, Makefile → REPORT non-standard patterns")
task(subagent_type="explore", load_skills=[], description="Find test patterns", run_in_background=true, prompt="Test patterns: FIND test configs/structure; ast-grep import-shape search + lsp_find_references on core modules to see what is covered → REPORT unique conventions")

# Measure project scale first
total_files=$(find . -type f -not -path '*/node_modules/*' -not -path '*/.git/*' | wc -l)
total_lines=$(find . -type f \\( -name "*.ts" -o -name "*.py" -o -name "*.go" \\) -not -path '*/node_modules/*' -exec wc -l {} + 2>/dev/null | tail -1 | awk '{print $1}')
large_files=$(find . -type f \\( -name "*.ts" -o -name "*.py" \\) -not -path '*/node_modules/*' -exec wc -l {} + 2>/dev/null | awk '$1 > 500 {count++} END {print count+0}')
max_depth=$(find . -type d -not -path '*/node_modules/*' -not -path '*/.git/*' | awk -F/ '{print NF}' | sort -rn | head -1)

// 500 files, 50k lines, depth 6, 15 large files → spawn 5+5+2+1 = 13 additional agents
task(subagent_type="explore", load_skills=[], description="Analyze large files", run_in_background=true, prompt="Large file analysis: FIND files >500 lines, REPORT complexity hotspots")
task(subagent_type="explore", load_skills=[], description="Explore deep modules", run_in_background=true, prompt="Deep modules at depth 4+: FIND hidden patterns, internal conventions")
task(subagent_type="explore", load_skills=[], description="Find shared utilities", run_in_background=true, prompt="Cross-cutting concerns: FIND shared utilities across directories")
// ... more based on calculation

Main Session: Concurrent Analysis

While background agents run, main session does:

1. Bash Structural Analysis

# Directory depth + file counts
find . -type d -not -path '*/\\.*' -not -path '*/node_modules/*' -not -path '*/venv/*' -not -path '*/dist/*' -not -path '*/build/*' | awk -F/ '{print NF-1}' | sort -n | uniq -c

# Files per directory (top 30)
find . -type f -not -path '*/\\.*' -not -path '*/node_modules/*' | sed 's|/[^/]*$||' | sort | uniq -c | sort -rn | head -30

# Code concentration by extension
find . -type f \\( -name "*.py" -o -name "*.ts" -o -name "*.tsx" -o -name "*.js" -o -name "*.go" -o -name "*.rs" \\) -not -path '*/node_modules/*' | sed 's|/[^/]*$||' | sort | uniq -c | sort -rn | head -20

# Existing AGENTS.md / CLAUDE.md
find . -type f \\( -name "AGENTS.md" -o -name "CLAUDE.md" \\) -not -path '*/node_modules/*' 2>/dev/null

2. Read Existing AGENTS.md

For each existing file found:
  Read(filePath=file)
  Extract: key insights, conventions, anti-patterns
  Store in EXISTING_AGENTS map

If --create-new: Read all existing first (preserve context) → then delete all → regenerate.

3. Code Map - drive LSP AND ast-grep (do NOT skip)

Highest-signal source for the CODE MAP and the Symbol/Export/Reference scoring rows. Complementary, not alternatives - run BOTH when present, alongside the explore agents.

LSP - check lsp_status; model-facing names are lsp_status/lsp_symbols/lsp_find_references/lsp_goto_definition (some harnesses drop the lsp_ prefix):

• lsp_symbols scope="document" on each entry point -> file outline.
• lsp_symbols scope="workspace", query by kind (class/interface/function) -> symbol inventory.
• lsp_find_references on top exports (line/character from the symbols result) -> reference centrality.

ast-grep - the ast-grep skill (sg with $VAR/$$$), or the ast_grep MCP (search/scan) when present; a first-class peer to LSP, NOT a last resort:

• import/export shapes -> module inventory; call shapes on core modules -> centrality + blast radius for the scoring matrix; class/function shapes -> symbol inventory.

rg covers text (string contents, comments, file names). For blast-radius/flow synthesis, fan out parallel explore agents armed with ast-grep. WHY: no symbol graph exists, so breadth comes from agents, not one graph query. Only if NEITHER LSP nor ast-grep resolves: mark centrality unmeasured in the CODE MAP.

Collect Background Results

// After main session analysis done, collect all task results
for each background task ID (`bg_...`): background_output(task_id="bg_...")

Merge: bash + LSP/ast-grep + existing + explore findings. Mark "discovery" as completed.

Phase 2: Scoring & Location Decision

Mark "scoring" as in_progress.

Scoring Matrix

Decision Rules

Output

AGENTS_LOCATIONS = [
  { path: ".", type: "root" },
  { path: "src/hooks", score: 18, reason: "high complexity" },
  { path: "src/api", score: 12, reason: "distinct domain" }
]

Mark "scoring" as completed.

Phase 3: Generate AGENTS.md

Mark "generate" as in_progress.

Root AGENTS.md (Full Treatment)

# PROJECT KNOWLEDGE BASE

**Generated:** {TIMESTAMP}
**Commit:** {SHORT_SHA}
**Branch:** {BRANCH}

## OVERVIEW
{1-2 sentences: what + core stack}

## STRUCTURE

{root}/ ├── {dir}/ # {non-obvious purpose only} └── {entry}

## WHERE TO LOOK
| Task | Location | Notes |
|------|----------|-------|

## CODE MAP
{From LSP/ast-grep - skip only if neither exists or project <10 files}

| Symbol | Type | Location | Refs | Role |
|--------|------|----------|------|------|

## CONVENTIONS
{ONLY deviations from standard}

## ANTI-PATTERNS (THIS PROJECT)
{Explicitly forbidden here}

## UNIQUE STYLES
{Project-specific}

## COMMANDS
```bash
{dev/test/build}

NOTES

{Gotchas}

**Quality gates**: 50-150 lines, no generic advice, no obvious info.

### Subdirectory AGENTS.md (Parallel)

Launch writing tasks for each location:

for loc in AGENTS_LOCATIONS (except root): task(category="writing", load_skills=[], run_in_background=false, description="Generate AGENTS.md", prompt=Generate AGENTS.md for: ${loc.path} - Reason: ${loc.reason} - 30-80 lines max - NEVER repeat parent content - Sections: OVERVIEW (1 line), STRUCTURE (if >5 subdirs), WHERE TO LOOK, CONVENTIONS (if different), ANTI-PATTERNS)

**Wait for all. Mark "generate" as completed.**

---

## Phase 4: Review & Deduplicate

**Mark "review" as in_progress.**

For each generated file:
- Remove generic advice
- Remove parent duplicates
- Trim to size limits
- Verify telegraphic style

**Mark "review" as completed.**

---

## Phase 5: Snapshot & Mode

Ask the user: **Local or committed?**

Capture the answer as `USER_MODE_CHOICE`. The explicit answer is authoritative:
- `local` keeps the generated guidance personal to this checkout.
- `committed` reruns the change through the `work-with-pr` skill so the generated guidance lands through a reviewed PR.
- If no explicit answer is available, tracked `AGENTS.md` status is the fallback.

Run these commands after the review is complete:

```bash
# Snapshot — create complete JSON with all fields, milliseconds timestamp
mkdir -p .omo
SHA=$(git rev-parse HEAD)
# Count tracked files (NUL byte counting, chunk-boundary safe)
FILES=$(git ls-files -z | node -e 'let c=0;process.stdin.on("data",d=>{for(let i=0;i<d.length;i++)if(d[i]===0)c++});process.stdin.on("end",()=>process.stdout.write(String(c)))')
LOC=$(git ls-files -z -- '*.ts' '*.tsx' '*.js' '*.jsx' '*.py' '*.go' '*.rs' '*.java' '*.kt' '*.swift' '*.rb' '*.php' '*.c' '*.cpp' '*.cs' '*.scala' '*.lua' '*.ex' '*.exs' '*.zig' '*.dart' | xargs -0 wc -l 2>/dev/null | tail -1 | awk '{print $1}')
NOW=$(node -e 'console.log(Date.now())')
USER_MODE_CHOICE="${USER_MODE_CHOICE:-}" && if [ "$USER_MODE_CHOICE" = "committed" ]; then MODE=committed; elif [ "$USER_MODE_CHOICE" = "local" ]; then MODE=local; elif git ls-files --error-unmatch AGENTS.md >/dev/null 2>&1; then MODE=committed; else MODE=local; fi
cat > .omo/init-deep.json <<EOF
{"commitSha":"$SHA","fileCount":$FILES,"loc":${LOC:-0},"timestamp":$NOW,"mode":"$MODE"}
EOF
# Local mode exclude — managed block (idempotent, never clobbers user lines)
if [ "$MODE" = "local" ]; then
  EXCLUDE=$(git rev-parse --git-path info/exclude)
  mkdir -p "$(dirname "$EXCLUDE")"
  if ! grep -q '# >>> omo-senpi init-deep local (managed)' "$EXCLUDE" 2>/dev/null; then
    cat >> "$EXCLUDE" <<'BLOCK'
# >>> omo-senpi init-deep local (managed)
# Do not edit this block; rerun init-deep or switch modes.
/.omo/init-deep.json
AGENTS.md
# <<< omo-senpi init-deep local (managed)
BLOCK
  fi
fi
# Nested AGENTS.md discovery
find . -name AGENTS.md -not -path '*/node_modules/*' -not -path '*/.git/*' -not -path '*/dist/*' -not -path '*/build/*'

When switching from local mode to committed mode, remove the managed block before rerunning through work-with-pr:

EXCLUDE=$(git rev-parse --git-path info/exclude) && sed -i.bak "/# >>> omo-senpi init-deep local (managed)/,/# <<< omo-senpi init-deep local (managed)/d" "$EXCLUDE" && rm -f "$EXCLUDE.bak"

USER_MODE_CHOICE=committed selects committed mode even when AGENTS.md is untracked. USER_MODE_CHOICE=local selects local mode even when AGENTS.md is tracked. .git/info/exclude cannot hide changes to an already tracked file, so explicit local mode on a tracked AGENTS.md is informational only and the file remains visible to git.

Final Report

=== init-deep Complete ===

Mode: {update | create-new}

Files:
  [OK] ./AGENTS.md (root, {N} lines)
  [OK] ./src/hooks/AGENTS.md ({N} lines)

Dirs Analyzed: {N}
AGENTS.md Created: {N}
AGENTS.md Updated: {N}

Hierarchy:
  ./AGENTS.md
  └── src/hooks/AGENTS.md

Anti-Patterns

• Static agent count: MUST vary agents based on project size/depth
• Sequential execution: MUST parallel (explore + LSP + ast-grep concurrent)
• Ignoring existing: ALWAYS read existing first, even with --create-new
• Over-documenting: Not every dir needs AGENTS.md
• Redundancy: Child never repeats parent
• Generic content: Remove anything that applies to ALL projects
• Verbose style: Telegraphic or die