The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from docs/review/issues.md to define tasks and remediation logic. While this presents an attack surface for instructions embedded in project documentation, the risk is effectively mitigated by mandatory human-in-the-loop (HITL) checkpoints. In Step 1 and Step 2, the user must explicitly confirm the range of issues to be addressed and the specific remediation plans proposed by the agent before any code modification occurs.
Ingestion points: docs/review/issues.md and project source code (read by subagents).
Boundary markers: Not explicitly present in prompt templates.
Capability inventory: File system modification, execution of build/test scripts, and shell operations (mkdir, mv).
Sanitization: None specified for the input data.
[COMMAND_EXECUTION]: The skill performs shell operations to manage project documentation and execute build processes.
Evidence: Step 4 uses shell commands date +%F, mkdir -p, and mv to archive review results.
Evidence: The skill executes build and test commands defined in the project's local configuration file docs/standards/tech-stack-rule.md during subagent verification and the final project build.
Context: These executions are standard for developer productivity tools and are confined to the local project environment.
[DYNAMIC_EXECUTION]: The skill is designed to modify the project's source code dynamically and then execute that code through its test and build suites.
Evidence: Subagents are instructed to modify real code and run verification tests using the project's specific tech-stack commands.
Context: This is the primary intended function of the skill and is protected by the requirement that subagents must ask for user confirmation if they encounter ambiguous situations or need to deviate from the agreed plan.

review-fix