The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from untrusted external files within a repository, which could contain malicious instructions designed to influence agent behavior.
Ingestion points: Files including plan.html, execution.md, executor.md, board-events.js, and board.jsonl are read into the context.
Boundary markers: The instructions do not define specific delimiters or instructions to the agent to disregard natural language commands embedded within these data files.
Capability inventory: The skill has the ability to read the file system, perform syntax checks using node --check, and execute validation scripts/tests.
Sanitization: There is no explicit sanitization or filtering of the content retrieved from the implementation records or plans.
[DYNAMIC_EXECUTION]: The skill instructs the agent to interact with and verify code by running shell commands and project-specific validation logic.
Evidence: Instructions include "Run node --check" on board-events.js and a requirement to "rerun focused validation rather than repeating a recorded success."
Context: While node --check is a syntax-only check, "focused validation" implies the execution of scripts or test suites present in the repository, which is a standard but noteworthy capability for an auditing skill.

qstack-plan-adherence-review