The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze potentially untrusted data from external sources, including application logs, traces, and API request/response bodies, which can serve as a vector for indirect instructions.
Ingestion points: SKILL.md (Workflow steps 2 and 4) and runner-specific references (e.g., references/omp.md, references/opencode.md) instruct the agent to read conversation context, provider streaming events, and server logs.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" markers to isolate ingested log or response data from the agent's primary task instructions.
Capability inventory: The skill utilizes file system read access to inspect project workspaces, log files, and configuration directories.
Sanitization: The skill includes robust sanitization instructions, explicitly mandating the redaction of API keys, authentication tokens, and sensitive headers in the Guardrails section of SKILL.md and the secret_rules across all runner reference files.
[NO_CODE]: The skill consists entirely of Markdown and YAML documentation and configuration. It does not include any executable scripts, binary files, or automated dependency installation routines.

model-detector