- Home
- Skills
- Code Quality & Review
- Code Intel: Semantic Navigation With LSP
More screenshots
Works with the AI Agents you already use
Code Intel: Semantic Navigation With LSP
Answers where code is defined, who calls it, and what it does in unfamiliar codebases, using targeted LSP queries instead of whole-file reads.
$5
Code Intel: Semantic Navigation With LSP
Example session with this skill installed
I just joined the team that owns the payments monorepo. Before I touch anything: what actually charges a card, and what breaks if the charge function's signature changes?
- Read your context and instructions
- Compiled the code intel
INTEL SESSION — produced by applying the skill (queries run from the repo root, TypeScript server verified against the workspace tsconfig first).
Shape first: symbol index over src/ shows billing/, checkout/, subscriptions/, refunds/, webhooks/ as the money-touching modules. The charge path lives in billing/.
Q: Where is a card actually charged?
Query: workspace/symbol "charge" → candidates ranked; definition on the one used by checkout → src/billing/charge.ts:41, createCharge. Hover: createCharge(orderId, paymentMethod, idempotencyKey) -> Promise<ChargeResult>; docstring: "the only path that talks to the payment provider; all callers must pass an idempotency key."
Q: Who calls it?
Query: references → 14 hits, grouped. Production: checkout/confirm.ts (the user-facing payment), subscriptions/renewal.ts (scheduled rebilling), refunds/partial.ts (negative-amount charges for partial refunds — flagged, because a signature change that assumes positive amounts breaks it silently). Tests: 6 hits across two files. Generated: 3 hits in the generated API client, traced back to the schema, not edited by hand.
Q: Is that all of them?
Cross-check down one rung: text search for the string "createCharge" finds a 15th reference the server did not return — the webhook handler dispatches retry events by name from a table in webhooks/handlers.ts. Labeled in the answer as dynamic dispatch: a rename refactor will not catch it.
Answer, with blast radius: changing the signature touches three production call sites (checkout, renewals, refunds), one string-keyed webhook dispatch, two test files, and the generated client via its schema. The refund caller's negative-amount use is the risk to design around. Every claim cited file:line; findings labeled by rung (14 LSP, 1 text search). Zero full-file reads — the only source pulled was charge.ts lines 41–96, to confirm the idempotency contract before reporting it.
Connects securely to your tools. The creator never sees your data.
About this skill
When you are dropped into unfamiliar code, you rarely need to read files — you need to answer five questions, and each one has a query cheaper than a read. Where is this defined? Who calls it? What does it take? What lives in this module? What changed around here? This skill turns those into a method: scope the codebase's shape first with a symbol index, classify the question, run the cheapest language-server query (definition, references, hover, workspace symbols), and cite every answer as file and line. Its core discipline is knowing when an empty result lies: an unindexed project and a missing symbol look identical, so it verifies the per-language setup before concluding anything is absent, then descends a fallback ladder — ctags, then text search, then a targeted read of the exact function — labeling which rung produced each finding. You get the five-question decision table, per-language setup notes for TypeScript, Python, Go, Rust, and C/C++, a question-log template, and an answer rubric that scores evidence, coverage, and economy. The output is answers with a blast radius, not source dumps.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Code Quality & Review.
- It answers where and who, not whether the code is correct — behavior that must be proven still needs running code and tests.
- Highly dynamic code (string dispatch, heavy metaprogramming) yields labeled inferences, not certainties.
- It needs at least one rung to stand on: no language server, no tag index, and no text search means no investigation.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
- Software
Python
Git