@devlensio/skill — DevLens Agent Skill

Teach your AI coding agent to understand your codebase — TypeScript, JavaScript, Python, Go, Rust, or Java (React/Next.js/Node, FastAPI/Flask/Django, Spring Boot, Gin/Echo/chi, axum/actix/rocket) — in one command.

What it does for you

Without DevLens, your AI agent reads files one at a time — it has no idea how components/classes/structs connect, what depends on what, or where security issues lurk. It burns tokens re-discovering the same connections every session.

With the DevLens Skill, your agent has a pre-built graph of every node in your codebase — with functional summaries, technical summaries, and security analysis. Type /devlens and get instant answers.

| Before (agent alone) | After (with DevLens Skill) | |----------------------|---------------------------| | Reads files one by one, 2,000+ tokens each | Queries the graph: ~50 tokens per node | | No idea how components connect | Sees every edge — IMPORTS, CALLS, PROP_PASS, navigations | | Misses architectural debt | Surfaces circular deps, god-files, coupling hotspots | | No per-component security insight | Every node has a security assessment | | Can't see big picture without reading everything | /devlens architecture — full overview instantly |

Real benchmarks — with DevLens vs without

Tested across 4 different LLMs (DeepSeek V4 Flash, GLM 5.2, Kimi K2.6, Qwen 3.6) on real-world tasks — architecture understanding, feature implementation, and bug finding — comparing the same model with and without DevLens.

Architecture understanding (full DevLens MCP)

| Metric | Without DevLens | With DevLens | Improvement | |--------|:--------------:|:------------:|:-----------:| | Avg cost per query | $0.163 | $0.075 | 54% cheaper | | Avg input tokens | 88,980 | 35,035 | 61% less | | Avg output tokens | 9,549 | 3,233 | 66% less | | Avg tool steps | 14.3 | 7.8 | 45% faster | | Structured architecture output | 50% | 100% | 2× more reliable | | Architectural debt discovered | 0% | 50% | Now discoverable | | Avg cache hit rate | 75.2% | 83.7% | +8.5pp |

Even the strongest model tested (DeepSeek V4 Flash) was 81% cheaper ($0.0035 vs $0.0185) and used 83% fewer input tokens with DevLens.

Feature implementation & bug finding (DeepSeek V4 Flash)

5 prompts across implementation and debugging tasks — DevLens graph context only (no per-node summaries).

| Task | Input tokens saved | Cache improvement | |------|:-----------------:|:-----------------:| | Continue Watching feature | 24% less input (56.8k vs 74.9k) | +5.2pp cache | | Rate Limiting feature | 22% less input (32.1k vs 41.1k) | +8.3pp cache | | Error Handling audit | (comparable due to simple task) | Comparable | | Profile Bug trace | 32% less input (36.9k vs 54.3k) | Similar |

Quality impact

When asked to explain a codebase's architecture:

| Capability | Without DevLens | With DevLens | |-----------|:--------------:|:------------:| | Produced structured output (diagrams, tables) | 50% | 100% | | Referenced specific graph metrics | 0% | 100% | | Identified architectural debt | 0% | 50% | | Named specific important files/modules | 75% | 100% |

(Benchmarked across DeepSeek V4 Flash, GLM 5.2, Kimi K2.6, and Qwen 3.6 — with and without DevLens.)

Install

npx @devlensio/skill install

This auto-detects which AI tools you're using and installs into each.

Options

| Flag | What it does | |:--|:--| | --harness=cursor | Force install for a specific tool | | --global | Install to home directory instead of project | | --force | Re-copy after an update |

Check your installation

npx @devlensio/skill check

Prerequisite

The DevLens CLI must be installed globally:

npm install -g @devlensio/cli

After installing, reload your AI tool and type /devlens:

New to a codebase?

| Command | What it does | Example | |---------|-------------|---------| | /devlens init | First run — connect MCP, configure provider, analyze repo | /devlens init → Sets up DevLens for the current repo | | /devlens architecture | Full system overview — stack, modules, all routes/stores, patterns | /devlens architecture → Returns: "Next.js 15 app with 12 routes, 4 state stores, 3 modules (auth, anime, user)..." | | /devlens explain [path] | Understand a specific module — callers, callees, summaries | /devlens explain src/auth/ → Explains the auth module: login, register, session management, and how they connect | | /devlens onboard | Generate a saved ONBOARDING.md for new devs | /devlens onboard → Writes ONBOARDING.md with setup steps, module map, key flows, and a reading path |

About to refactor?

| Command | What it does | Example | |---------|-------------|---------| | /devlens impact <symbol> | Blast radius — what breaks if you change this? | /devlens impact loginUser → "Changing loginUser affects 14 dependents: LoginPage, AuthGuard, Navbar (via auth store), ProfilePage..." | | /devlens tech-debt | Circular deps, coupling hotspots, god-files | /devlens tech-debt → "3 circular dependency groups, Navbar has 28 dependents (highest fan-in), utils.ts has 12 functions (candidate for splitting)" | | /devlens guard [target] | Warns if you're about to edit a critical/high-blast-radius node | /devlens guard (no arg) → Checks uncommitted changes; warns if modifying authMiddleware (score 9.1, affects 22 dependents) |

Before shipping?

| Command | What it does | Example | |---------|-------------|---------| | /devlens security-analysis [level] | Prioritized security findings with exploit notes | /devlens security-analysis high → "2 high-severity issues: SQL injection in loginUser (reach: 14 nodes), missing CSRF in deleteAccount" | | /devlens changes [range] | Explain recent work — what changed and its impact | /devlens changes yesterday → "3 files changed: added analytics tracking, fixed rate-limiting bug in API handler" | | /devlens diagram [type] | Mermaid diagrams — architecture, clusters, flows, deps | /devlens diagram architecture → Generates a layered Mermaid diagram with module map, route flows, and external dependencies |

Need to find something?

| Command | What it does | Example | |---------|-------------|---------| | /devlens find <name> | Locate any component, class, function, struct, or route | /devlens find Button → "Found 3 matches: src/ui/Button.tsx (COMPONENT, score 6.2), src/ui/IconButton.tsx (COMPONENT, score 4.1), tests/Button.test.tsx (TEST)" | | /devlens summary <target> | On-demand technical + business + security summaries | /devlens summary security Navbar → "Security: Navbar reads user role from auth store — no privilege escalation risk, but exposes admin links" |

Supported tools

| Tool | Project install dir | Global install dir | | :-- | :-- | :-- | | Claude Code | .claude/skills/devlens/ | ~/.claude/skills/devlens/ | | Cursor | .cursor/skills/devlens/ | ~/.cursor/skills/devlens/ | | Kilo Code | .kilo/skills/devlens/ | ~/.kilocode/skills/devlens/ | | opencode | .opencode/skills/devlens/ | ~/.config/opencode/skills/devlens/ | | pi | .agents/skills/devlens/ | ~/.pi/agent/skills/devlens/ |

Without --harness, the installer auto-detects which tools are in use from their marker directories (.claude, .cursor, .kilo, .opencode, .agents).

Alternative: Claude Code plugin

/plugin marketplace add devlensio/devlensOSS
/plugin install devlens@devlensio

Commands are namespaced as /devlens:devlens <subcommand>.

How it works

1. DevLens analyzes your codebase — walks the AST, builds a graph of every node + typed edges, scores by importance, generates AI summaries
2. The Skill teaches your agent — when to query the graph, how to keep output token-cheap, how to produce thorough architecture/security/impact reports
3. You type /devlens <command> — agent queries the graph via MCP, returns structured results

License

AGPL-3.0. Part of the DevLens project.