The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local repository scripts using the Bash tool to validate the integrity of generated files. Specifically, it runs python3 scripts/check-skill-integrity.py --verbose and ./scripts/check-surface-sync.sh. These are internal development tools used for linting and consistency checks within the skill's own environment.
[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that ingests user-provided intent and existing session stubs to generate new executable instruction files (SKILL.md). This creates an attack surface where untrusted input is converted into agent instructions.
Ingestion points: User fuzzy intent gathered during the interview phase and session discovery stubs processed via the --from-learn flag.
Boundary markers: The skill utilizes a Phase 1 interview process and a confirmation "design brief" to structure the input before file generation.
Capability inventory: Write (to create persistent skill files), Bash (to execute local validation scripts), and Read (to analyze existing repo skills).
Sanitization: The skill enforces strict naming conventions (kebab-case) and a predefined structural template, though it relies on the model to safely interpret user intent into instructions.

new-skill