The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute a variety of system commands and development utilities (such as go run, cargo test, npm test, rg, vulture, and staticcheck) to gather evidence, run tests, and identify dead code within a target backend repository.
[EXTERNAL_DOWNLOADS]: In the references/dead-code.md file, the skill suggests using npx knip, which can download and execute packages from the public NPM registry, an established well-known service.
[PROMPT_INJECTION]: The skill processes external, untrusted content from the repositories it audits, creating an indirect prompt injection surface.
Ingestion points: Target directory files (source code, README files, OpenAPI specifications) and outputs from discovery commands (references/evidence.md).
Boundary markers: The skill does not use specific delimiters but relies on its own instructional flow to contextualize data.
Capability inventory: The agent possesses capabilities for shell command execution, file system modification, and deletion of source code (via the --sweep parameter).
Sanitization: To mitigate risks, references/verbs/study.md includes a mandatory 'Safety/source check' rule that instructs the agent to extract factual data only and explicitly ignore any embedded instructions encountered in the studied source material.

anvil