index

name	index
description	Use when indexing codebase components or external resources for future specs.
argument-hint	[--path=dir] [--type=types] [--exclude=patterns] [--dry-run] [--force] [--changed] [--quick]
allowed-tools	Read Write Agent Bash AskUserQuestion Glob Grep WebFetch ListMcpResourcesTool
disable-model-invocation	true

Index Command

You are running the codebase indexing command. This scans the repository to generate component specs in specs/.index/.

Checklist

Use native task UI tracking if TaskCreate is available; otherwise keep this checklist inline:

• Parse arguments and validate
• Generate project topology snapshot
• Pre-scan interview (skip if --quick)
• Scan codebase components
• Fetch external resources (URLs, MCP, skills)
• Generate component and external specs
• Build index summary
• Post-scan review (skip if --quick)

Step 1: Parse Arguments

Parse from $ARGUMENTS:

Option	Format	Default	Description
--path	--path=<dir>	Project root	Limit scan to directory
--type	--type=<types>	All	Comma-separated: controllers,services,models,helpers,migrations
--exclude	--exclude=<patterns>	See below	Comma-separated exclude patterns
--dry-run	Flag	false	Preview without writing
--force	Flag	false	Regenerate all (ignore existing)
--changed	Flag	false	Only git-changed files
--quick	Flag	false	Skip interviews, batch only

Default excludes: node_modules, vendor, dist, build, .git, __pycache__, *.test.*, *.spec.*, *_test.*, test/, tests/, spec/, specs/

Validation: --force + --changed together is an error. --path must exist. --type values must be valid. --changed requires git (git rev-parse --git-dir).

Step 2: Project Topology Snapshot

Generate the cheap topology cache before component scanning:

mkdir -p specs/.index
node "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/lib/project-topology.mjs" \
  --goal "$ARGUMENTS" > specs/.index/project-topology.json
node "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/lib/project-topology.mjs" \
  --goal "$ARGUMENTS" \
  --format context-map > specs/.index/context-map.md

Use this snapshot to decide which roots are relevant. If project-topology.json contains missingRoots, stop indexing and show the accessFix value exactly.

Step 3: Pre-Scan Interview

**Skip if --quick flag is detected.**

Ask via AskUserQuestion:

#	Question	Key
1	External documentation URLs to index?	externalUrls
2	MCP servers or skills to document?	externalTools
3	Specific directories to focus on?	focusAreas
4	Code areas lacking comments needing extra attention?	sparseAreas

Store responses in ./specs/.index/.index-state.json under interviewResponses.

Step 4: Component Scanner

Detection Patterns

Category	Glob Pattern	File Pattern
Controllers	**/controllers/**/*.{ts,js,py,go}	*controller*, *Controller*
Services	**/services/**/*.{ts,js,py,go}	*service*, *Service*
Models	**/models/**/*.{ts,js,py,go}	*model*, *Model*
Helpers	**/helpers/**/*.{ts,js,py,go}	*helper*, *Helper*, *util*
Migrations	**/migrations/**/*.{ts,js,sql}	*migration*

Scanning Process

1. For each category (filtered by --type if set), run Glob with --path as base. If no --path is provided, scan each accessible code root from project-topology.json. Filter out excludes.
2. If --changed: filter to git diff --name-only HEAD results only
3. If not --force and spec exists: compare SHA-256 hash (first 8 chars via shasum -a 256), skip if unchanged
4. For each matched file, extract metadata using language-appropriate Grep patterns:
   • Exports: export (const|function|class|interface|type) \w+ (TS/JS), ^def \w+\( / ^class \w+ (Python), ^func \w+\( (Go)
   • Methods: Class method signatures per language
   • Dependencies: import/require/from statements per language
5. For large codebases (>100 files): process in batches of 50, update state after each batch

Step 5: External Resource Fetcher

Process resources from interview responses. Classify each: URL (http(s)://), MCP server (mcp-*/mcp_*), or Skill (/ prefix or contains "skill").

URL Processing

For each URL in externalUrls:

1. WebFetch with prompt to extract key sections, API endpoints, config options
2. Generate external spec from templates/external-spec.md
3. Write to specs/.index/external/url-<sanitized-name>.md
4. On timeout/error: warn and skip, continue processing

MCP Server Processing

For each MCP server in externalTools:

1. ListMcpResourcesTool to discover tools and resources
2. Document each tool/resource in external spec
3. Write to specs/.index/external/mcp-<server-name>.md

Skill Processing

For each skill in externalTools:

1. Read plugin manifests: plugins/*/.claude-plugin/plugin.json
2. Extract commands, agents, hooks from manifest
3. Write to specs/.index/external/skill-<plugin>-<command>.md

File Naming

Pattern: <type>-<sanitized-name>.md. Sanitize: lowercase, replace non-alphanumeric with hyphens, collapse consecutive hyphens, max 50 chars.

Step 6: Generate Specs

Ensure directories: mkdir -p specs/.index/components specs/.index/external

Component Specs

For each scanned component:

1. If --dry-run: collect for preview, do NOT write
2. Load templates/component-spec.md, fill template variables:
   • {{SOURCE_PATH}}, {{CONTENT_HASH}}, {{CATEGORY}}, {{TIMESTAMP}}
   • {{COMPONENT_NAME}}, {{AUTO_GENERATED_SUMMARY}}, {{EXPORTS}}, {{METHODS}}
   • {{DEPENDENCIES}}, {{KEYWORDS}}, {{RELATED_FILES}}
3. Write to specs/.index/components/<category>-<basename>.md
4. Update hash in .index-state.json

Naming: <singular-category>-<lowercase-basename>.md (e.g., controller-users.md)

External Specs

Load templates/external-spec.md, fill {{SOURCE_TYPE}}, {{SOURCE_ID}}, {{FETCH_TIMESTAMP}}, {{RESOURCE_NAME}}, {{CONTENT_SUMMARY}}, {{SECTIONS}}, {{KEYWORDS}}, {{RELATED_COMPONENTS}}.

Index Summary

Load templates/index-summary.md, aggregate counts by category, list all components and externals. Write to specs/.index/index.md.

Dry-Run Mode

If --dry-run: write nothing. Display preview table:

Dry Run - Would generate:
| File | Category | Source | Status |
|------|----------|--------|--------|
Total: N files

Step 7: Post-Scan Review

**Skip if --quick flag is detected.**

Ask via AskUserQuestion:

#	Question	Key
1	Found {{count}} components. Seem complete?	componentCount
2	External resources look correct?	externalResources
3	Any areas to re-scan or adjust?	adjustments

Handle feedback: re-scan missing areas, re-filter if too many, re-process changed externals.

Step 8: Update State and Output

Update specs/.index/.index-state.json with lastIndexed, componentCount, externalCount, categories, hashes.

Display:

Index complete for '{{project_name}}'.
Output: specs/.index/

| Category | Count |
|----------|-------|
| Controllers | {{count}} |
| Services | {{count}} |
| Models | {{count}} |
| Helpers | {{count}} |
| Migrations | {{count}} |
| External Resources | {{count}} |
| **Total** | **{{total}}** |

Next: Run /curdx-flow:start to create specs that reference indexed components.

Error Handling

Scenario	Action
No components found	Warn, suggest --path=src/ or broader patterns. Still create specs/.index/
External URL unreachable	Warn and skip, continue others
MCP server unavailable	Warn and skip, continue others
Git unavailable + --changed	Fatal error, suggest --force
Permission denied on file	Skip file with warning, continue
Hash unchanged (no --force)	Skip silently, count in summary
Template missing	Use inline minimal fallback
State file corrupted	Reset state, full scan
Monorepo detected	Preserve package context in paths, group by package in summary

Never abort the entire index for recoverable errors. Show error/warning summary at end.