The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as an orchestrator that ingests untrusted data from multiple sources which could contain malicious instructions.
Ingestion points: The skill processes user-supplied featureDescription and techStack inputs, along with existing repository files (e.g., AGENTS.md, skills/<projectSlug>/SKILL.md) and codebase files mapped during the orientation phase.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when interpolating these external inputs into its planning phases.
Capability inventory: The workflow has high-impact capabilities including shell command execution (npm, bun, pytest, rspec) and persistent file system write access for scaffolding new skills and tasks.
Sanitization: There is no evidence of sanitization or filtering applied to the feature descriptions or existing project documentation before they are used to generate the task specification in Phase 7.
[COMMAND_EXECUTION]: The skill is designed to execute various local shell commands as part of the development lifecycle.
Evidence: Phase 10 and Phase 12 explicitly call for running test runners and build tools, specifically mentioning npm run test, rspec-rails, vitest, pytest, and bun run build.
[DYNAMIC_EXECUTION]: The workflow involves the runtime generation and registration of executable code structures.
Evidence: Phase 8 ('BLUEPRINT CONTRACT') scaffolds a new executable workflow contract at a computed path (skills/<projectSlug>/workflows/<feature-slug>/SKILL.md) and wires it into the project's routing matrix for future execution.

embed-aihero-radioactive