The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute a sequence of build and verification tools including npx, go, and python3. It specifically invokes local management scripts: harness/scripts/code-logger.py, harness/scripts/qc-regression.py, and harness/validators/task_lifecycle.py to enforce workflow gates.
[DYNAMIC_EXECUTION]: The skill dynamically determines which commands to run based on the presence of project manifest files (e.g., using npm test if package.json is found). It also assembles shell commands by interpolating variables like task IDs and commit hashes into script arguments for lifecycle management.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local markdown files in the llmwiki/wiki/sources/draft/ directory. It extracts task identifiers from YAML frontmatter to update task states via external scripts. This behavior creates an attack surface where maliciously crafted local drafts could attempt to influence the execution parameters of the lifecycle scripts.
Ingestion points: Frontmatter and content of markdown files in llmwiki/wiki/sources/draft/.
Boundary markers: Relies on file system paths and YAML structure; no explicit instruction-ignoring delimiters are mentioned for the draft content.
Capability inventory: Shell execution of repository build tools and local Python management scripts across the pre-commit and post-commit phases.
Sanitization: The instructions do not specify validation or escaping for the extracted task IDs before they are passed to the shell-based script invocations.

verify-before-commit