The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the local repository to drive its implementation logic. While this is a standard operational surface, it creates a vector for indirect instructions.
Ingestion points: The agent reads task files (docs/monorail/<feature>/tasks/), specifications (spec.md), context documents (CONTEXT.md), and ADRs as seen in the 'Process' and 'Exploration' sections of SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard embedded directions within these project files.
Capability inventory: The skill utilizes Git (commit, worktree), file system writes, sub-agent spawning, and shell command execution for testing and typechecking.
Sanitization: No sanitization or validation of the ingested markdown content is performed before it influences the agent's actions.
[DYNAMIC_EXECUTION]: The skill's primary purpose involves generating code and executing it through testing frameworks.
Evidence: The skill drives TDD at specified seams, runs typechecks, and executes 'relevant tests' to verify implementation, as described in Step 6 and Step 7 of the 'Process' section in SKILL.md.

rail-build