The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes shell scripts (scripts/assess-code.sh and scripts/assess-tests.sh) that perform static analysis using system utilities such as grep, awk, and wc. These scripts are intended to be executed by the agent within the local development environment to inspect code quality.
[DYNAMIC_EXECUTION]: The scripts/assess-tests.sh script executes npm test and npm run test:coverage. This involves executing code defined within the project's package.json and test files. If the skill is used to review untrusted or malicious code, this execution path provides a vector for local code execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted code provided by users. It lacks explicit boundary markers or instructions to ignore instructions embedded within the code (e.g., in comments), creating a surface for indirect prompt injection.
Ingestion points: Code snippets or file paths passed to the qe-code-reviewer agent for analysis (referenced in SKILL.md).
Boundary markers: Absent; the instructions do not define clear delimiters for the code being reviewed.
Capability inventory: Execution of shell scripts and project-level build tools (npm).
Sanitization: No evidence of input sanitization or filtering for the code being processed.
[PROMPT_INJECTION]: The SKILL.md file contains a 'Minimum Findings Enforcement' section that mandates the agent find at least 3 weighted findings. This instruction overrides the model's objective judgment and may force the agent to fabricate issues or escalate the severity of minor concerns to meet the quota.

brutal-honesty-review