The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes repository documentation which acts as an ingestion point for potentially untrusted data. Ingestion points: Repository tree, README files, manifests, and existing context files. Boundary markers: None identified in the workflow instructions. Capability inventory: Reading project documentation and writing/patching CONTEXT.md files. Sanitization: None identified; content is categorized based on domain taxonomy. This finding is assessed as safe as the ingestion is central to the skill's purpose and is mitigated by human verification steps.\n- [SAFE]: No evidence of obfuscation, hardcoded credentials, or unauthorized network operations was found. The agent configuration includes safety-oriented policies such as 'disable-model-invocation' and 'allow_implicit_invocation: false', which limit autonomous activity.

bootstrap-context