Agent-Skills.md

Agent Skills: Scan Codebase

Perform a fresh-context scan of code and docs that turns verification observations and missing-test gaps into findings. Accept optional `path` and `depth` parameters and default to `path=infer`, `depth=deep`. Confirm effective variables before starting.

UncategorizedID: agenta-ai/agenta/scan-codebase

Repository

agenta-ai/agenta
Agenta-AILicense: NOASSERTION
4,261558

Install this agent skill to your local

pnpm dlx add-skill https://github.com/Agenta-AI/agenta/tree/HEAD/agents/skills/scan-codebase

Skill Files

Browse the full folder contents for scan-codebase.

Download Skill

Loading file tree…

agents/skills/scan-codebase/SKILL.md

Skill Metadata

Name
scan-codebase
Description
Perform a fresh-context scan of code and docs that turns verification observations and missing-test gaps into findings. Accept optional `path` and `depth` parameters and default to `path=infer`, `depth=deep`. Confirm effective variables before starting.

Scan Codebase

Read these shared references when needed:

  • ../shared/references/findings.schema.md
  • ../shared/references/findings.lifecycle.md

Role

Run a fresh review pass from code and docs into findings.

  • This skill is verification-oriented.
  • It may surface missing tests or coverage gaps as findings.
  • It does not rely on existing findings as the starting point.
  • It does not replace test-codebase; if runtime validation is needed, hand off there.

Depth

Accept a depth parameter from the prompt:

  • shallow
  • deep

Default:

  • depth=deep

Path Input

Accept an optional path from the prompt.

  • If path is provided, use it as the local design or findings folder.
  • If path is omitted, infer it from the branch, subsystem, or matching docs and state the inferred value before starting.

Default:

  • path=infer

Workflow

  1. Establish scan scope. Confirm the effective variables first:

    • path
    • depth
    • branch, base, subsystem, and any requested focus areas

  2. Start from fresh context. Read current code, docs, tests-as-code, routes, schemas, migrations, and adjacent design material before looking at existing findings.

  3. Produce findings from review. Surface correctness, consistency, completeness, soundness, functionality, security, performance, compatibility, migration, and missing-test findings when supported by evidence.

  4. Update the active findings record. Use path/findings.md.

  5. Hand off when execution is needed. Use triage-findings for discussion and planning, test-codebase for validation, and resolve-findings for implementation.

Rules

  • Bias toward deep inspection, not prior-thread anchoring.
  • Do not silently preload old findings into the initial reasoning path.
  • Do not present test execution as completed if you only inferred missing coverage from code.