The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub pull requests (titles, descriptions, and code diffs) to perform reviews. \n
Ingestion points: Data is fetched from GitHub via gh pr view and gh pr diff in scripts/fetch-pr-data.sh, scripts/review-pr.md, and scripts/run-pr-checks.py. \n
Boundary markers: Absent. The ingested content is processed directly in the agent context without specific delimiters or warnings to ignore embedded instructions. \n
Capability inventory: The skill can execute shell commands (gh, ruff, eslint), read/write files, and perform network operations via the GitHub CLI. \n
Sanitization: No sanitization or filtering is performed on the externally sourced PR content beyond standard string parsing. \n- [COMMAND_EXECUTION]: The skill executes multiple external commands to fetch data and perform static analysis. \n
Evidence: scripts/run-pr-checks.py uses subprocess.run to invoke gh, ruff, and npx eslint. scripts/run-lint-check.sh and scripts/fetch-pr-data.sh utilize shell commands to interact with gh, biome, and eslint. \n
Context: These are standard developer tools integrated for the skill's intended purpose. Python subprocess calls use list-based arguments, which is a secure pattern that mitigates shell injection. \n- [DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context markers to incorporate live project state during initialization. \n
Evidence: scripts/review-pr.md contains dynamic context commands (!) such as !gh pr list and !gh pr list --head "$(git branch --show-current)". \n
Context: These commands are used to provide the agent with immediate context regarding the local git state and recent pull requests. This behavior is consistent with the skill's developer-focused utility.

code-review-playbook