The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of simulation requests and answers through the peridigm_start_context and peridigm_answer_context functions. This data is used to populate simulation specifications which are then passed to execution tools like pd-agent. Maliciously crafted input could attempt to influence the agent's interpretation of results or its tool-calling behavior.
Ingestion points: SKILL.md (Clarify and confirm section) and references/capabilities.md (Workflow boundary).
Boundary markers: The skill implements a confirm_spec_sha256 hash for finalized specifications, but lacks explicit boundary delimiters for the initial raw user input.
Capability inventory: Subprocess execution via pd-agent CLI, launching detached workers with peridigm_submit, and file system operations via peridigm_repair and peridigm_collect.
Sanitization: The skill incorporates automated checks such as peridigm_validate and peridigm_doctor to verify the integrity of simulation decks before they are processed.

peridigm-evidence-fracture