agentleFS
vibe-cokitvibe-cokit/claude-code/skills/context-engine/context-engine-exploring/SKILL.md

context-engine-exploring

Skill1 starsChanged 4 months agoApache-2.0View on GitHub

Use when the user asks how code works, wants to understand architecture, find where something lives, or explore an unfamiliar codebase via the Context Engine MCP. Examples: \"How does X work?\", \"Where is the auth logic?\", \"What calls this function?\"

The file

Files
SKILL.md

Exploring Codebases with Context Engine

When to Use

  • "How does authentication work?"
  • "What's the project structure / where do I find X?"
  • "Show me the main components of feature Y"
  • "What calls this function?" (via call-graph annotations)
  • Understanding code you haven't seen before

Workflow

1. list-repos                                              → get repoId, confirm status=indexed
2. codebase-retrieval({repoId, information_request: "<concept you want to understand>"})
                                                           → ranked snippets across the whole repo
3. Follow [callers:…] / [calls:…] annotations to navigate the call graph
4. file-retrieval({repoId, file_path, information_request}) → zoom into a specific file
5. Read <file>:<lines>                                     → exact content before any edit

If list-repos shows the index is stale (lastCommitSha ≠ local HEAD) → see context-engine-sync.

Checklist

- [ ] list-repos → repoId + freshness check
- [ ] codebase-retrieval with a NATURAL-LANGUAGE description of the concept (not an identifier)
- [ ] Review the ranked snippets and their file#line ranges
- [ ] Follow [calls:…] outward / [callers:…] inward to map the flow
- [ ] file-retrieval to drill into the most relevant file
- [ ] Read the source at the returned line ranges for implementation detail

Tools

codebase-retrieval — the primary tool. Describe WHAT you want in natural language:

codebase-retrieval({repoId, information_request: "single-app aggregate Hono bootstrap and route mounting"})
→ core/apps/iam/src/app.ts#L1-87 [calls: createServiceApp, registerRoutes, …]
→ apps/models-router/apps/api/src/app.ts#L1-133 [callers:3]
→ core/packages/sdk-server/src/hono/factory.ts#L1-86 [callers: buildApp, buildIamApp +8 more]
  • Good query: "Where is the database connection pool initialised?"
  • Bad query: "find definition of class Foo" → use grep for exact identifiers
  • limit (default 30, max 50) widens/narrows the result set.

file-retrieval — when you know the file but not the line range:

file-retrieval({repoId, file_path: "core/apps/iam/src/app.ts", information_request: "where publicPaths for STS are declared", top_k: 5})
→ snippets with line numbers → then Read to confirm

Query Strategy

You want to…information_request phrasing
Understand a feature end-to-end"how <feature> works from request to response"
Locate a responsibility"where <responsibility, e.g. rate limiting> is done"
Find the entry/bootstrap"service bootstrap / app composition / main entry"
Find config/contract for a domain"<domain> Zod schema / contract definition"

Example: "How does the IAM PDP authorize requests?"

1. list-repos → repoId=be1730bd-…, status=indexed
2. codebase-retrieval({repoId, information_request: "IAM PDP authorize decision evaluation flow"})
   → core/apps/iam/src/api/authz.routes.ts, services/authz/… (PDP)
3. Follow [calls:…] into the policy evaluation + decision cache symbols
4. file-retrieval on authz.routes.ts for the exact handler
5. Read the handler + evaluator for implementation details

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.