The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing existing 'dirty' repositories to perform rehabilitation and audits. Malicious instructions hidden within the source code, comments, or documentation of these repositories could potentially influence the agent's reporting or planning decisions.
Ingestion points: The scripts/audit_web_app_contract.py utility and the agent's inventory process (using Read and Bash) ingest various files including source code (.tsx, .js), configuration files, and documentation from the target project.
Boundary markers: The process requires a 'pessimistic plan' stored on disk and a 'visual decision review' for human approval, which serves as a major control to prevent unauthorized autonomous actions.
Capability inventory: The skill is granted access to Bash, Write, Edit, WebFetch, and WebSearch tools, allowing it to modify the environment and fetch external references.
Sanitization: The provided audit script uses regular expressions to detect secrets and design violations, but there is no general sanitization of the natural language content extracted from the repository before it is processed by the agent.

ideal-web-app-builder