Index of every page on this site: llms.txt
All skills
vercel-labs avatar

/deepsec

@9444b89
by Vercel Labsvercel-labs/deepsec8.5k stars
521
Run it yourself

Run deepsec, an AI-powered cyber-security vulnerability scanner. Activates when the user invokes /deepsec, asks to run deepsec, or wants to scan their repo, branch, or uncommitted changes for vulnerabilities.

Use this Skill: https://skilld.dev/gh/vercel-labs/deepsec/deepsec

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈54 tokens for metadata: the name and description. ≈1.2k when used: this file. ≈313k more on demand in 78 files.

Description uses 2.7% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈218 characters, or 2.7%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

/deepsec — scan this repository with deepsec

deepsec is an AI-powered vulnerability scanner.

  • Modern AI models are great at security code review.
  • Most review solutions only run on pull requests.
  • That means most legacy code was never reviewed, and your code from 6 months ago was reviewed by older models.
  • Instead deepsec does security review on ALL of your existing code using scaled VM fanout.
  • Vercel's deepsec is open source, runs in your own infrastructure, and supports strong agent sandboxing.

A fast regex scan flags candidate files, then AI agents investigate each candidate in depth and record findings with severity ratings; findings can later be revalidated, triaged, and exported. Everything it adds to a repository lives in a single .deepsec/ workspace (config, installed package, per-project data). Because processing runs real AI agents, it costs money in proportion to how much code it investigates.

Follow this runbook when the user invokes /deepsec or asks for a scan.

1. Ask for scope first

Ask the user which scope to process, using a structured question tool if you have one (AskUserQuestion in Claude Code), otherwise plain text:

  • Uncommitted changes — working tree + untracked files
  • Diff to main — changes vs origin/main (use main if there is no origin remote)
  • Entire codebase — warn that this is the expensive option: AI processing investigates every candidate file and can cost real money on a large repository

Ask before doing anything else so the rest of the flow can run unattended.

2. Detect onboarding state

From the repository root:

  • No .deepsec/deepsec.config.ts → not onboarded. Do step 3 in full.
  • .deepsec/ exists but .deepsec/node_modules/deepsec is missing, or a previous setup was interrupted → re-run the init command from step 3; it resumes from checkpoints and repairs the install rather than starting over.
  • Otherwise → onboarded; skip to step 4.

3. Onboard without full processing

Onboarding normally ends with an AI processing pass over the whole repository. Since the user already chose a scope, stop setup after the coverage phase — that still includes install, login, threat model, matcher generation, and the final regex scan, but skips the full-repo AI process phase. The scoped processing happens in step 4 instead.

From the repository root, inspect the read-only plan, then run setup:

npx -y deepsec init --plan --output json
npx -y deepsec init --yes --through coverage --output jsonl

If npx fails with EBADDEVENGINES, the repository's devEngines.packageManager requires another package manager. Run the same commands with pnpm dlx deepsec (or that package manager's equivalent) in place of npx -y deepsec.

Parse every output line as JSON. On a needs_input event, show the supplied message and actions to the user rather than inventing remediation. In particular, VERCEL_AUTH_REQUIRED normally asks the user to run npx vercel login; after they do, follow the returned link action from inside .deepsec (use npx vercel link when the user needs to choose a project), then re-run the same init command. Exit code 2 means input is needed; exit code 3 means a cost/duration boundary stopped the resumable run — re-running the same command resumes it. Never expose credential values, bypass --yes prompts on the user's behalf beyond the flag itself, or launch an interactive login yourself.

4. Run the scoped processing

Run from inside .deepsec/ (the config loader only finds deepsec.config.ts in the current directory or its ancestors; after step 3, npx deepsec resolves to the copy installed there):

Scope Command
Uncommitted changes cd .deepsec && npx deepsec process --diff-working
Diff to main cd .deepsec && npx deepsec process --diff origin/main
Entire codebase, right after step 3 cd .deepsec && npx deepsec process (the final scan from setup already produced the candidate set)
Entire codebase, previously onboarded cd .deepsec && npx deepsec scan && npx deepsec process

5. Interpret results

  • Direct-mode (--diff*) exit codes: 0 = no net-new findings, 1 = at least one net-new finding (not an error), anything else = runtime error. Pre-existing findings on touched files are excluded from the gate.
  • Summarize any findings for the user, then offer follow-ups (all from inside .deepsec/): npx deepsec report, npx deepsec revalidate, and npx deepsec export --format md-dir --out ./findings.

Going deeper

After onboarding, full documentation ships with the installed package at .deepsec/node_modules/deepsec/dist/docs/ — getting-started.md, reviewing-changes.md (direct mode, exit codes, CI gating), configuration.md, models.md, and more. Read the relevant doc before varying the commands above; flags and defaults change between releases.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

2 warnings1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    This skill is a security scanner that uses AI agents to perform in-depth vulnerability research. It possesses powerful capabilities, including the ability to execute shell commands and access the local filesystem to facilitate scanning. While these features are central to its function and are supported by robust documentation and sandboxing options, they represent a significant security surface that should be reviewed before use, especially when analyzing untrusted codebases.

  • Socket1mo

    14 alerts: gptSecurity, gptAnomaly

  • Snyk1mo

    Risk: unknown · No issues

Signed by skilld at 9444b89. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 weeks ago

README badge

README badge for vercel-labs/deepsec