The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly directs the agent to execute shell commands, lint configuration files, and compile code snippets found within user-provided manuscripts to verify their accuracy (documented in SKILL.md and references/core/rules.md).
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface. It is designed to ingest and process technical content from external, untrusted manuscripts which could contain malicious commands or instructions intended to hijack the agent's behavior during the auditing process.
Ingestion points: Technical artifacts such as commands, code, and links extracted from the manuscript in Step 2 of workflows/verify-technical-manuscript.md.
Boundary markers: The skill does not define specific delimiters or instructions for the agent to treat manuscript content as untrusted data rather than instructions during the verification phase.
Capability inventory: The agent is granted permission to run commands and code within a test environment as part of the verification logic described in references/core/rules.md.
Sanitization: The skill includes significant safety constraints in references/core/rules.md, such as Rule 4 (Prioritize Reader Risk) and the Exceptions section, which forbids executing destructive, privileged, or costly commands unless the environment is explicitly disposable.

technical-manuscript-verification