The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various local Go-based CLI tools (e.g., assess, compound, events, journal) found in the user's project directory. These tools are used for state management, activity logging, and event tracking. The commands use a project slug identifier which is derived from the local environment or user selection.
[PROMPT_INJECTION]: The skill processes PRDs and value propositions from the local filesystem, creating a surface for indirect prompt injection.
Ingestion points: Markdown files located in .agents/products/prds/ and .agents/products/value-propositions/.
Boundary markers: None; the instructions direct the agent to read the files directly into context.
Capability inventory: Local shell command execution (Go scripts, Git), file system writes, and Stitch MCP tool invocation.
Sanitization: None; the skill relies on the underlying agent's standard safety filters when evaluating the ingested text. This surface area is considered safe as it is intrinsic to the skill's primary document-review function.
[DATA_EXFILTRATION]: The skill utilizes an event emission system to log progress. This involves passing JSON payloads to a local execution script (events emit). The data consists of phase markers and summary counts of PRD changes, which is standard for developer workflow tracking.

product-vision