The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses markdown files in the docs/plans/ directory to define agent objectives and constraints. This creates an indirect ingestion surface where repository content influences agent logic.\n
Ingestion points: scripts/check-complete.mjs and scripts/create-goal-scratchpad.mjs parse plan files and project metadata (e.g., AGENTS.md).\n
Boundary markers: The skill enforces a strict markdown schema with mandatory sections like Objective:, Constraints:, and Boundaries:, providing clear structural delimiters.\n
Capability inventory: The agent utilizes standard capabilities for command execution, file modification, and browser interaction to satisfy the plan's requirements.\n
Sanitization: check-complete.mjs performs mechanical validation to ensure required fields contain concrete content (non-placeholder values), serving as a basic integrity check.\n- [DATA_EXFILTRATION]: The skill searches project files to identify Linear workspace slugs to generate documentation links.\n
Evidence: scripts/create-goal-scratchpad.mjs recursively scans docs/plans/ and reads files like README.md to find linear.app references.\n
Context: This behavior is limited to constructing informational links to a well-known project management service and does not involve exfiltrating sensitive data to unauthorized third parties.

autogoal