The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled shell script scripts/mine_review_comments.sh using the Bash tool. This script utilizes gh and jq to interface with the local repository's GitHub data. It performs network operations specifically to GitHub's official API to retrieve pull request metadata and review text.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from external sources (GitHub PR comments) which could contain instructions designed to manipulate the agent's behavior.\n
Ingestion points: The body field of review comments and review verdicts retrieved in scripts/mine_review_comments.sh.\n
Boundary markers: The instructions do not specify delimiters or warnings for the agent to ignore instructions embedded within the mined review comments.\n
Capability inventory: The skill possesses Write, Edit, and Bash tools, which could be misused if the agent blindly follows instructions found in review text.\n
Sanitization: There is no evidence of text sanitization or filtering to remove potentially malicious commands or instructions from the review bodies.\n
Sanitization (Mitigation): A safety control is present in the Confirm one at a time phase, where the AskUserQuestion tool is used to obtain explicit human approval for each proposed convention before it is written to the filesystem.

generate-conventions