The Agent Skills Directory

[COMMAND_EXECUTION]: The verification harness component (verify.mts) is designed to execute arbitrary shell commands defined in Markdown 'recipe' files to verify project features.
Evidence: The runShown function in references/harness/verify.mts uses node:child_process.spawn to execute commands via sh -c inside a process-monitoring wrapper.
Evidence: Example recipes like references/harness/recipes/create-note.md contain Run: lines that execute local scripts (e.g., scripts/control-notes recipe ...).
[DYNAMIC_EXECUTION]: The skill facilitates the dynamic assembly and execution of shell commands at runtime based on the contents of the project's verification recipes.
Evidence: The harness parses Markdown recipes to extract Run: commands and then executes them using the VERIFY_COMMAND environment variable.
[INDIRECT_PROMPT_INJECTION]: The project 'mining' functionality (Step 5) requires the agent to ingest and analyze untrusted data from the repository's history, which could contain malicious instructions.
Ingestion points: As defined in references/MINING.md, the skill reads Pull Request bodies, issue discussions, git commit messages, and documentation files to reconstruct project rationale.
Capability inventory: The skill has the ability to write project configuration files (CONTEXT.md, NORTH_STAR.md) and the verification harness can execute shell commands.
Boundary markers: SKILL.md instructs the agent to interpret only top-level options as flags and treat other mentions as topic text to prevent parameter injection.
Sanitization: references/harness/verify.mts includes a localPaths function that uses a regex to scrub local user home directories and temporary paths from the evidence logs generated by command execution.

setup-project