The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest untrusted data from the repository (specifications, tracker items, and code) to construct the handoff prompt. * Ingestion points: External files defined in the 'Resolved sources' (Section 2) and 'Verification' (Section 6) phases. * Boundary markers: The skill includes explicit instructions to draw scope boundaries (Section 4) and follows grounding rules to point to paths rather than paraphrasing content. * Capability inventory: The skill uses tools to read files and refers to 'project-board' and 'handoff-report' tools (Sections 8 and 9) to save or transmit the generated content. * Sanitization: There are no explicit instructions to sanitize the content of ingested files for hidden or malicious instructions before including them in the sub-agent's prompt.
[COMMAND_EXECUTION]: The skill instructs the agent to extract build, lint, and test commands from the repository's manifests (Section 6) for inclusion in the handoff prompt. While intended for validation, this involves identifying shell commands defined in external project configuration files for use by sub-agents.

handoff-prompt