The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted input from pull request bodies, comments, and review threads. It specifically instructs the agent to fetch and recurse through linked references and external URLs found in the PR context. This architecture creates a surface for indirect prompt injection where an attacker could influence the agent's planning or execution by embedding instructions in comments or linked documents.\n
Ingestion points: SKILL.md (Workflow Step 1) instructs the agent to gather PR title, body, comments, review threads, and linked refs/URLs.\n
Boundary markers: The instructions do not define any delimiters or specific instructions for the agent to isolate untrusted data from its system instructions.\n
Capability inventory: The skill uses gh and git for repository management and executes verification commands during the fix process as described in SKILL.md and references/tdd-cycle.md.\n
Sanitization: There is no evidence of sanitization or validation logic for the content fetched from external sources before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill relies on the GitHub CLI (gh) and git to perform its primary tasks, including metadata retrieval, log fetching, and thread resolution. It also executes arbitrary commands during the verification phase of the TDD cycle and as part of the fix confirmation process.

snap-resolve