The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill outlines a workflow that ingests untrusted external data, such as bug reports and feature requests (via the /triage skill) and code diffs (via /code-review). This creates a vulnerability surface where malicious instructions embedded in external inputs could attempt to influence the agent's behavior in later steps. Ingestion points: SKILL.md identifies /triage and /code-review as points where untrusted external content enters the agent's context. Boundary markers: The documentation discusses context management and phase boundaries but does not specify delimiters or instructions to ignore embedded commands within the untrusted data. Capability inventory: The workflow includes powerful capabilities such as writing files (/implement), executing tests (/tdd), and generating bash scripts (/wizard). Sanitization: No explicit sanitization or validation of external content is described.
[DYNAMIC_EXECUTION]: The skill mentions the /wizard tool, which generates interactive bash scripts at runtime for provisioning and setup tasks. Generating executable scripts based on variable agent context involves dynamic code creation.

ask-matt