The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard local commands including grep, glob, and git log to analyze the project's structure and history. It also utilizes project-defined linting and build scripts (e.g., npm run lint, pytest) to verify that code removals do not break functionality. All commands are confined to the local project environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted project source code during its detection phase, creating a surface for indirect prompt injection. However, this risk is effectively mitigated by the skill's design:
- Ingestion points: The skill reads local source files using glob and grep patterns defined in Phase 2.
- Boundary markers: The skill maintains strict boundaries by producing a structured report at .claude/dead-code-report.md and employing an explicit AskUserQuestion confirmation step before any action.
- Capability inventory: The agent possesses read/write/delete capabilities on the local filesystem and the ability to execute local build/lint tools, but has no network access permissions.
- Sanitization: While no automated sanitization of code content is performed, the 'human-in-the-loop' requirement for cleanup approval serves as the primary security control.
[SAFE]: No malicious patterns such as credential harvesting, network exfiltration, persistence mechanisms, or privilege escalation were detected. The skill's author metadata and resource patterns are consistent with legitimate vendor functionality.

dead-code-sweep