The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the bash tool to perform workspace management and process monitoring. Evidence includes:
File system operations: rm -f, mkdir -p, and mktemp to manage temporary review artifacts.
Process monitoring: Use of grep, jq, tail, and tac to parse NDJSON logs and determine the state of background tasks.
Execution of npx repomix to serialize the codebase for LLM consumption.
[DYNAMIC_EXECUTION]: The skill employs dynamic script and prompt generation techniques to coordinate subagents:
It constructs a complex orchestrator prompt at runtime, writes it to a file (review-prompt.txt), and then executes it by reading the file back into a shell command (opencode run ... "$(cat "$PROMPT_FILE")").
It generates bash assembly scripts on-the-fly to combine per-area review files into a final report.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because it processes the entire contents of a repository. A malicious file within the codebase could potentially influence the behavior of the review agents.
Ingestion points: The repomix.xml file created in Step 2, which contains the full content of the target repository.
Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore embedded instructions within the codebase.
Capability inventory: The agents have access to bash for file operations and task for spawning additional subagents.
Sanitization: None; the repository content is packed and processed in its raw form.
[EXTERNAL_DOWNLOADS]: The skill uses npx repomix, which fetches the repomix utility from the NPM registry at runtime if it is not already cached locally.

review-deep