Tarosec Install Tarosec

Audit

Review a local project for security risks in source code, dependencies, CI/CD, and agent integrations.

Quick start

  1. Install Tarosec and open your project in your coding agent.
  2. Paste the request below into the agent chat. Replace ./project with your project path, or use . when your agent is already in the project root.
  3. Review the findings and coverage gaps before asking the agent to make changes. Audit runs in read-only mode by default.
Review a project
$tarosec ./project
Review this project for security risks. Start with a quick audit of authentication and configuration. Cite files and line numbers, and list anything you could not verify.

What to provide

Use a local checkout you are authorized to review. Tell the agent which files, components, or recent changes matter most. If you have a previous finding, advisory, or scanner report, give the agent its local path.

Choose an audit mode

Quick
Get a first pass on the highest-risk areas.
Deep
Follow a sensitive operation through the code.
Dependencies
Check whether a reported issue affects the versions in your lockfile.
Agents
Review skills, hooks, prompts, and MCP tool permissions.
Verify
Check whether a change addresses a specific finding.

Quick audit

Use the default quick mode when you want to know where to start. It looks first at exposed entry points, privileged actions, possible secret exposure, and risky configuration. Ask the agent to name the areas it leaves for a later review.

Example prompt
$tarosec-audit
Run a quick audit of ./project. Focus on public routes, admin permissions, and deployment configuration. Give me the highest-priority findings with file and line references, and list areas you deferred.

Deep audit

Use deep when you need evidence about a particular feature or boundary. Give the agent a narrow scope, such as who may read customer exports. It follows the operation from the entry point to the protected data and checks what actually enforces access.

Example prompt
$tarosec-audit
Run a deep audit of the customer export feature in ./project. Trace each entry point to its authorization checks and the data it reads. Check whether the tests assert who can access an export. Cite the code and list any boundaries you could not verify.

Dependency audit

Use dependencies to review an existing advisory or scanner report. Provide the lockfile so the agent can check the exact resolved version and trace how the package enters your project. The report should distinguish an affected version from evidence that the application uses the affected code.

Example prompt
$tarosec-audit
Use dependencies mode on ./project. Compare ./project/package-lock.json with my saved dependency report at ./dependency-report.json. For each match, show the resolved version and dependency path. Flag stale or unverified entries. Keep the review offline.

Agent and MCP audit

Use agents for a local skill, hook, agent configuration, or MCP server definition. Ask the reviewer to trace what data reaches a tool, what action the tool can take, and which permission checks protect that action. The reviewer inspects the supplied files as data.

Example prompt
$tarosec-audit
Use agents mode to review the skill and MCP configuration files I supplied from ./project. Trace which inputs can reach tools that read or share data. Cite the permission checks you can verify and flag unverified controls. Review the files without activating the skill or running hooks.

Verify a fix

Use verify after a fix. Supply the original finding and its ID, the updated source, and any saved regression results. Ask the agent to revisit the same code path so you can compare the result with the original finding.

Example prompt
$tarosec-audit
Use verify mode for finding AUDIT-001 in ./findings.md. Review the current code in ./project and my saved regression results. Keep the original finding ID. Explain whether the fix addresses the original issue, with evidence and any remaining uncertainty.

Read the results

  • Evidence shows where the issue appears and which conditions must be present for it to matter.
  • Severity and confidence tell you how serious the impact could be and how well the evidence supports the conclusion.
  • Remediation and regression checks describe the proposed fix and how to verify it.
  • Coverage shows what the agent inspected, what it could not verify, and which checks it actually ran. Secret values are redacted.

Replace the sample paths and finding ID with your own. If evidence is missing, ask the agent to name the specific file, report, or test result it needs.