The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill coordinates workflows involving external, untrusted data sources which present a vector for indirect prompt injection.
Ingestion points: SKILL.md and references/behavior-fixtures.md identify logs, issues, web pages, and tool output as data the agent must process. references/behavior-fixtures.md includes simulated build logs with malicious instructions.
Boundary markers: Instructions in SKILL.md act as boundary markers by directing the agent to treat external content as evidence, not instructions.
Capability inventory: The skill coordinates multi-stage software delivery including implementation, verification, and deployment actions.
Sanitization: SKILL.md and references/work-records.md provide logic for protecting sensitive credentials and ignoring instructions embedded in external artifacts.
[EXTERNAL_DOWNLOADS]: The skill provides links to documentation and guidelines from well-known technology organizations.
Technical references in references/sources.md include links to official blogs and repositories from Anthropic, OpenAI, and Humanlayer for workflow guidance.

agentic-sdlc