The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as pull request diffs, reporter evidence, and contributor implementation notes. This creates a vulnerability where adversarial instructions within a PR or bug report could influence the agent's logic during the solution gating process. Ingestion occurs in SKILL.md and references/candidate-audit.md. While isolation techniques are used, explicit prompt sanitization or boundary markers for this data are not documented. The agent has capabilities to execute commands and write to files, increasing the impact of a successful injection.
[COMMAND_EXECUTION]: The skill mandates the execution of a local script, scripts/resolve-source-root.mjs, in Step 7 to manage decision logging. This script is an external dependency not contained within the skill files.
[DYNAMIC_EXECUTION]: The runtime execution of the unbundled script scripts/resolve-source-root.mjs allows the skill to perform operations defined outside of its static instructions.

solution-gate