The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions direct the agent to clone an external repository from GitHub (https://github.com/antbotlab/blueprint.git). While the instructions utilize a pinned commit hash for reproducibility and recommend a review process, the repository belongs to an organization that is not on the verified list.
[COMMAND_EXECUTION]: The skill provides a sequence of shell commands for the agent to perform during the setup phase, including mktemp, git clone, git checkout, git ls-files, and git status.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves scanning the active codebase and project memory to generate plans. This creates a potential ingestion surface for malicious instructions hidden in the files being analyzed.
Ingestion points: Reads local codebase files and project memory during the 'Research' phase (SKILL.md).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing external file content.
Capability inventory: The skill is designed to run 'pre-flight checks' and generate execution plans, which may involve further tool or shell operations.
Sanitization: No sanitization or validation of the scanned codebase content is documented in the instructions.
[METADATA_POISONING]: The skill manifest contains a risk: critical field. While this appears to be an author-defined safety warning rather than a deceptive tactic, it is a non-standard use of metadata fields that could influence an agent's perception of its own safety profile.

blueprint