The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell-like commands such as ls, grep, and cat through an exec interface to interact with the local project structure. These operations are scoped to the proposals/ and decisions/ directories and are used for routine tasks like checking for existing documents and determining the next sequence number for new files.- [INDIRECT_PROMPT_INJECTION]: This skill possesses an attack surface for indirect prompt injection as it ingests content from external markdown files within the repository using search and read tools.
Ingestion points: Content is read from the proposals/ and decisions/ directories in SKILL.md.
Boundary markers: The skill does not explicitly define character-based delimiters for external content but mandates human review gates (Step 1 and Step 9) before advancing the state of any document.
Capability inventory: The skill has the capability to execute shell commands (exec), search project history (search), and modify the filesystem (write, edit).
Sanitization: There is no automated sanitization of the markdown content read from the files; however, the skill relies on the agent's internal logic and human oversight to evaluate the ingested text.

frame-a-proposal