The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages a multi-step workflow ($scout, $investigate, $plan, etc.) and uses the spawn_agent tool to orchestrate sub-agents for complex documentation tasks.
[PROMPT_INJECTION]: The instructions include directives such as 'Ignore Claude-specific mode-switch instructions' and 'Strict execution contract.' These are evaluated as framework-specific steering used to maintain consistency across platform versions rather than attempts to bypass safety filters.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect injection as it processes user prompts as workflow context.
Ingestion points: The user's prompt is passed as the primary context for the workflow-feature-spec workflow in SKILL.md.
Boundary markers: The skill relies on 8-section tech-free template enforcement and mandatory success criteria definitions to bound the AI's behavior.
Capability inventory: Capabilities include sub-agent spawning via spawn_agent and filesystem access for reading and updating project documentation in the docs/ directory.
Sanitization: The workflow mandates evidence-based claims with specific file:line citations and 'tech-free' output constraints to ground the generated content in repository evidence.

workflow-feature-spec