The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains instructions intended to override standard agent workflows by directing it to bypass approval gates during code migrations. The text states: "A request to migrate, consolidate or enforce typography authorizes the implementation steps; do not add another approval gate."
[INDIRECT_PROMPT_INJECTION]: The workflow relies on the agent reading and interpreting untrusted source code from the user's repository (Step 2) to categorize UI roles. This information is then used to perform automated migrations (Step 7), creating a path for malicious content within the source code to influence the agent's behavior.
Ingestion points: Source code files scanned during Step 2 of the workflow.
Boundary markers: None provided in the instructions for the agent when interpreting code context or "jobs."
Capability inventory: Includes file system modification (migration steps) and command line execution.
Sanitization: None specified for handling or escaping content retrieved from the scanned repository.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a provided Python script (scripts/census-typography.py) against the local filesystem to audit typography usage. While the script uses standard libraries, the execution of local scripts on a repository is a sensitive operation.

typecase