Indexes files and maps semantic code relationships to improve graph-based navigation and AI context.
Install
mkdir -p .claude/skills/yams && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/2538" && unzip -o skill.zip -d .claude/skills/yams && rm skill.zipInstalls to .claude/skills/yams
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Code indexing, exact/semantic search, graph-assisted code navigation, and project memoryKey capabilities
- →Indexes codebase for semantic search
- →Maps relationships between code symbols
- →Attaches metadata (owner/phase/task) to work items
- →Visualizes dependency graphs
- →Auto-updates index on file changes via watcher
How it works
Maintains a local knowledge graph database that maps code tokens and their context, updated by a background daemon.
Inputs & outputs
When to use yams
- →Search code with semantic context
- →Explore relationships between code symbols
- →Track project memory across tasks
- →Index new files for rapid lookup
About this skill
YAMS Skill (agent.md)
Quick Reference
# Status & Health
yams status # Check daemon and index status
yams daemon start # Start background daemon
yams doctor # Diagnose issues
# Indexing
yams add <file> # Index single file
yams add . -r --include "*.py" # Index directory recursively
yams watch # Auto-index on file changes
# Search (use grep first, search for semantic)
yams grep "pattern" --cwd . # Code pattern search scoped to current project
yams grep -e "--flag" --cwd . # Explicit pattern for leading '-' text
yams grep -g "*.cpp" "TODO" # rg-style glob filtering
yams grep --minimal "TODO" # Compact grep-style output
yams search "query" # Semantic/hybrid search
# Hydrate/export a selected search result
yams cat --hash <hash> # Inspect saved content on stdout
yams get --hash <hash> -o <path> # Export only when a file copy is needed
# Graph
yams graph --explore <query> # Agent context: symbols, relationships, snippets
yams graph --name <file> # Raw file relationships
yams graph --list-types # List node types with counts
yams graph --relations # List relation types with counts
yams graph --search "pattern" # Search nodes by label
# graph owns path/tree/topology inspection; retired tree is not top-level
# Agent storage
yams list --format json # Scriptable list output
yams list --show-metadata # Include metadata for work item
Agent Memory Workflow
YAMS is the single source of truth for agent memory and work item.
Required Metadata (Task Tracking)
Attach metadata to every yams add.
task- short task slug (e.g.,list-json-refresh)phase-start|checkpoint|completeowner- agent or authorsource-code|note|decision|research
Index Project Files
# Index specific file types
yams add . -r --include "*.ts,*.tsx,*.js"
# Index with exclusions
yams add . -r --include "*.py" --exclude "venv/**,__pycache__/**"
# Index with metadata for tracking
yams add src/ -r --metadata "task=list-json-refresh,phase=checkpoint,owner=codex,source=code"
Auto-Index with Watch
yams watch # Start watching current directory
yams watch --interval 2000 # Custom interval (ms)
yams watch --stop # Stop watching
Verify Indexing
yams status # Shows indexed file count
yams list --limit 10 # Recent indexed files
Search Patterns
Decision Tree
- Code patterns →
yams grep(fast, regex/literal; use--cwd .for repo scoping) - Semantic/concept →
yams search(embeddings/hybrid) - Codebase shape / blast radius →
yams graph --explorefrom a search/grep hit - Path/tree/topology inspection → use
yams graph, not retired top-leveltree - No results from grep → Try
yams search, then followgraph_explore_hintwhen present - Saved memory content → search/list discovers candidates;
yams cat --hash <hash>hydrates the selected artifact
Search, list, and grep snippets are routing context, not the complete saved memory. Hydrate the most relevant one to three note/decision/research/evidence hits before using them. For code hits, prefer graph narrowing plus a targeted local read instead of reading every result in full.
grep (Code Search)
# Exact pattern
yams grep "function authenticate"
# Regex pattern
yams grep "async.*await.*fetch"
# With context lines
yams grep "TODO" -A 2 -B 2
# Filter by extension
yams grep "import" --ext py
# rg-style glob filter (repeatable)
yams grep -g "src/**/*.cpp" "TODO"
# Compact grep-style output (default is richer agent context)
yams grep --minimal "TODO" -g "src/**/*.cpp"
# Scope to current working directory or an explicit directory
yams grep "TODO" --cwd .
yams grep "TODO" --cwd src/daemon
# Literal text (no regex)
yams grep "user?.name" -F
# Per-file match cap: --limit is an alias for -m/--max-count, not global
# Use search/list --limit for global result caps.
yams grep "TODO" -m 5
# Pattern starts with '-': use -- or explicit -e/--regexp
yams grep -- "--tags|foo" --include="docs/**/*.md"
yams grep --regexp "--tags|foo" -g "docs/**/*.md"
search (Semantic Search)
# Concept search
yams search "error handling patterns"
# Hybrid search (default)
yams search "authentication flow" --type hybrid
# Limit results
yams search "database connection" --limit 5
# Filter by file type
yams search "API endpoint" --ext ts
search (Metadata-Only)
# Force metadata/FTS path for structured metadata
yams search "task=example-task" --type keyword --limit 10
# Unique task selection (avoid collisions)
yams search "task=example-task" --type keyword --limit 20
# 2) List all used task values with counts
# Tag filters (tags are stored as metadata keys: tag:<name>)
yams search "plan" --type keyword --tags plan --limit 10
yams search "tagged logic" --type keyword --tags plan --limit 20
Agent Storage
Store Research
# Index documentation
curl -s "https://docs.example.com/api" | yams add - --name "api-docs.md" \
--metadata "task=docs-cache,phase=checkpoint,owner=codex,source=research"
# Store with metadata
yams add notes.md --metadata "task=research-auth,phase=checkpoint,owner=codex,source=research"
Store Decisions
# Pipe decision record
echo "## Decision: Use JWT for auth
### Context
Need stateless authentication for microservices.
### Decision
JWT with RS256, 15min expiry, refresh tokens.
### Rationale
Stateless, scalable, industry standard.
" | yams add - --name "decision-jwt-auth.md" \
--metadata "task=auth-decision,phase=checkpoint,owner=codex,source=decision"
Retrieve Knowledge
# Discover related decisions; results emit an exact cat command.
yams search "authentication decision" --limit 10
# Hydrate a selected saved-memory artifact before relying on it.
yams cat --hash <hash-from-result>
# Export only when a filesystem copy is needed.
yams get --hash <hash-from-result> -o <path>
# Find by exact metadata.
yams list --format json --show-metadata \
--metadata "owner=codex" --metadata "task=example-task" \
--metadata "source=decision" --limit 10
# Metadata + tags are separate in JSON output
yams list --format json --show-metadata \
| jq '.documents[] | {name,metadata,tags}'
For CLI workflows, cat is the inspection/hydration hop and get -o is the
export hop. For MCP workflows, chain search to get with
include_content: true, as shown below.
Session Management
Create Work Sessions
# Start named session
yams session start --name "feature-auth"
# List sessions
yams session ls
# Switch session
yams session use "feature-auth"
# Show current session
yams session show --json
Session Scope
# Add files to session scope
yams session add --path "src/auth/**"
# Warm session cache (faster searches)
yams session warm --limit 100
# Search within session
yams search "login" --session
Session Lifecycle
# Save session state
yams session save
# Load previous session
yams session load --name "feature-auth"
# Clear session cache
yams session clear
# End session
yams session close
Graph Queries
Use graph after search/grep finds a likely entry point. Graph answers "what is connected to this?" and should guide local reads, not replace them.
Agent Graph Context
# Preferred follow-up after search/grep hints: ranked symbols + line-numbered snippets
yams graph --explore "authenticateUser" --max-files 3
# Explore a file path when the result path is more useful than a symbol name
yams graph --explore src/auth/login.ts --max-files 8
# JSON for tool consumers
yams graph --explore "RequestHandler" --json
Notes:
yams searchandyams grepresults may emitgraph_explore_hint; run that exact command before broad local search.--exploreis budgeted for agents: entry symbols, related files, relationship summaries, and line-numbered snippets.- If
--explorefails or looks stale, fall back to raw traversal plus local reads.
Raw Graph Structure
# List available node types and relation types
yams graph --list-types
yams graph --relations
# Search nodes by label pattern (wildcards: * any chars, ? single char)
yams graph --search "*Controller*"
yams graph --search "auth*"
yams graph --search "handle?Request"
# List scoped node types
yams graph --list-type function --scope-cwd --limit 50
File / Symbol Relationships
# Show file dependencies and symbols
yams graph --name src/auth/login.ts --depth 2 --limit 50
# Filter by relation type when doing blast-radius review
yams graph --name src/main.ts --relation includes --depth 1
yams graph --node-key "func:authenticate" --relation calls --depth 2
# Output as JSON or DOT
yams graph --name src/auth/login.ts --format json
yams graph --name src/auth/login.ts --format dot > graph.dot
Common relations: calls, includes, contains, defined_in, located_in, has_version, semantic_neighbor.
Topology Navigation
# Find subsystem clusters and their medoid/bridge/core files
yams graph --topology-snapshots
yams graph --topology-clusters
yams graph --cluster <cluster-id>
Use topology when entering an unfamiliar subsystem. Start with medoids for representative files, bridges for cross-subsystem coupling, and core files for local implementation detail.
MCP Integration
YAMS exposes tools via Model Context Protocol for programmatic access.
Start MCP Server
yams serve # Start MCP server (quiet mode)
yams serve --verbose # With logging
Available MCP Tools (Code Mode)
The MCP server exposes a small composite tool surface.
| Tool | Purpose |
|---|---|
query | Read-only pipeline: search, grep, |
Content truncated.
When not to use it
- →Searching for extremely small, non-indexed snippets
- →Working in non-persistent ephemeral environments
Prerequisites
Limitations
- →Initial indexing can be time-consuming
- →Memory usage grows with project size
- →Requires daemon to be running
How it compares
It uses semantic memory to navigate project relationships instead of simple regex searching.
Compared to similar skills
yams side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| yams (this skill) | 1 | 2mo | Review | Intermediate |
| scientific-brainstorming | 37 | 7mo | No flags | Intermediate |
| webclaw | 1 | 4mo | Review | Intermediate |
| graphify-knowledge-graph | 0 | 4mo | Review | Intermediate |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
scientific-brainstorming
davila7
Research ideation partner. Generate hypotheses, explore interdisciplinary connections, challenge assumptions, develop methodologies, identify research gaps, for creative scientific problem-solving.
webclaw
0xmassi
Web extraction engine with antibot bypass. Scrape, crawl, extract, summarize, search, map, diff, monitor, research, and analyze any URL — including Cloudflare-protected sites. Use when you need reliable web content, the built-in web_fetch fails, or you need structured data extraction from web pages.
graphify-knowledge-graph
Aradotso
> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.
markitdown
K-Dense-AI
Convert various file formats (PDF, Office documents, images, audio, web content, structured data) to Markdown optimized for LLM processing. Use when converting documents to markdown, extracting text from PDFs/Office files, transcribing audio, performing OCR on images, extracting YouTube transcripts, or processing batches of files. Supports 20+ formats including DOCX, XLSX, PPTX, PDF, HTML, EPUB, CSV, JSON, images with OCR, and audio with transcription.
whisper
davila7
OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual ASR.
devtu-optimize-skills
mims-harvard
Optimize ToolUniverse skills for better report quality, evidence handling, and user experience. Apply patterns like tool verification, foundation data layers, disambiguation-first, evidence grading, quantified completeness, and report-only output. Use when reviewing skills, improving existing skills, or creating new ToolUniverse research skills.