The Agent Skills Directory

[PROMPT_INJECTION]: The skill reads external data from work/ directories to perform its assessment. This creates an indirect prompt injection surface where instructions could be embedded in the data files. However, given the specific review-centric purpose and lack of dangerous execution sinks, this is considered a standard operational characteristic.
Ingestion points: work/<slug>/README.md and work/<slug>/data/ (SKILL.md)
Boundary markers: None specified.
Capability inventory: Edit(work/**), Bash(ls), Bash(jq) (SKILL.md frontmatter)
Sanitization: Not present.
[COMMAND_EXECUTION]: Utilizes ls and jq to inspect artifacts and process metadata. These commands are used for analysis and are restricted to the work directory context.

review-work