The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically GitHub pull request comments, reviews, and Notion ticket descriptions. This data is used to drive automated actions like code patching and commit generation, creating a surface for indirect prompt injection.\n
Ingestion points: The skill reads PR feedback using gh pr view and fetches Notion ticket content via Notion tools in SKILL.md (Step 5) and BABYSITTING.md (On each wake).\n
Boundary markers: There are no explicit delimiters or instructions to the agent to isolate potential command-like text within the ingested comments.\n
Capability inventory: The skill has broad capabilities including shell access (Bash), file modification (Edit, MultiEdit), and source control operations (git push, gh pr create).\n
Sanitization: No sanitization or verification of the external text is performed before the agent is instructed to patch locally or answer the feedback.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform git operations, interact with the GitHub CLI, and run background monitoring loops.\n
Evidence: SKILL.md uses shell commands for syncing branches and pushing code. BABYSITTING.md implements a merge waiter background process using a bash for loop and sleep commands.

create-pr