The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external PR comments.\n
Ingestion points: Fetches PR comment content from GitHub using the cache-read-comment.sh script in SKILL.md (Step 1).\n
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when passing comment content to background tasks (Step 3.4).\n
Capability inventory: The skill utilizes Task, Bash, Edit, and Write tools, enabling it to modify code and execute commands.\n
Sanitization: No explicit sanitization or validation of the external comment content is performed before interpolation into sub-task prompts.\n
Mitigation: The process is entirely interactive, requiring the user to decide on the fix and approve the action for each individual item, providing a robust human-in-the-loop check.\n- [COMMAND_EXECUTION]: The skill executes local shell scripts to manage the PR workflow and synchronize data with GitHub.\n
It calls environment-provided scripts like get-pr-number.sh, cache-read-comment.sh, and cache-write-comment.sh from the ${CLAUDE_PLUGIN_ROOT}/scripts/ directory.\n
It employs standard utilities such as jq for JSON manipulation and mktemp with trap for secure temporary file handling during metadata updates.

pr-review-resolver