The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to fetch remote pull request branches using git fetch and execute their dependency installation and test scripts.
Evidence: Commands such as npm ci && npm test, pip install -r requirements.txt && pytest, cargo test, and go test are executed within a worktree containing the PR code (SKILL.md, Step 6).
Risk: An attacker could submit a malicious pull request that executes arbitrary code when the validator is run.
[DYNAMIC_EXECUTION]: The skill uses python3 -c to execute inline Python scripts that process JSON data received via a pipe from curl.
Evidence: curl -sSL "https://registry.npmjs.org/<pkg>" | python3 -c '...' and similar patterns for PyPI and GitHub APIs (SKILL.md, Step 4a, 4b).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Git commit messages and external release notes to identify package changes.
Ingestion points: git log output, curl responses from registries and GitHub (SKILL.md).
Boundary markers: None specified for the data processed by the Python scripts or the report generation.
Capability inventory: Shell execution, network access, file system read/write.
Sanitization: Partial, as the data is parsed via json.load() in Python scripts, but the extracted text is then used to generate a natural language report.
[EXTERNAL_DOWNLOADS]: Fetches package information and release notes from official registries and GitHub.
Sources: registry.npmjs.org, pypi.org, api.github.com, and raw.githubusercontent.com (SKILL.md, Step 4).
Note: These are well-known services and the downloads are documented neutrally as part of the skill's functional requirements.
[COMMAND_EXECUTION]: Relies on a wide range of shell commands for repository manipulation, dependency management, and running tests.
Evidence: git fetch, git worktree, npm, pip, cargo, go, mvn, and grep are used throughout the workflow.

dependabot-validator