The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes a review-findings.md ledger containing a recheck cell which is treated as a runnable verification method. An attacker who can influence the contents of this ledger could potentially execute arbitrary commands in the agent's environment.
Ingestion points: The recheck cell within the review-findings.md file (referenced in references/FOLD_POLICY.md).
Boundary markers: The skill lacks explicit instructions to treat the data as untrusted or to ignore embedded instructions within the ledger data.
Capability inventory: The skill uses shell execution for verification gates and reproducers, Git operations (commit, push), and file system writes.
Sanitization: There is no evidence of sanitization or validation logic for the commands extracted from the ledger before they are executed.
[COMMAND_EXECUTION]: The skill is designed to run shell commands for the project's "verification gate" and to "materialize" reproducers. This is a powerful capability that, while intended, increases the impact of any injection vulnerabilities.

fold-findings