lc-again
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads problem status information from
README.md, which could potentially act as a vector for indirect instructions if the file content contains malicious data designed to influence the agent. - Ingestion points: Reads status cells from
README.md(Step 1) to parse the current verdict and pass count. - Boundary markers: No explicit delimiters or instructions are used to separate the parsed file content from the agent's task context.
- Capability inventory: The skill possesses the ability to execute local scripts via the
Bashtool and modify the file system using theEdittool. - Sanitization: The logic relies on matching existing idioms and does not appear to perform sanitization on the notes or metadata carried within the README cells.
- [COMMAND_EXECUTION]: The skill triggers the execution of local repository scripts to synchronize data and verify the impact of changes.
- Evidence: Specifically invokes
bash script/get_again_problems.sh,python3 script/suggest_review.py, andpython3 script/eval_lc_readiness.py. These are internal tools provided by the repository author to maintain the local environment.
Audit Metadata