The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process external documents provided by the user, such as previous grant proposals and reviewer comments. This creates an attack surface where malicious instructions hidden in those documents could potentially influence the agent's behavior.
Ingestion points: The skill ingests untrusted content in Phase 3 (Quality Review), Phase 4 (Simulated Review), and the 'Adapt from Previous Proposal' workflow as described in SKILL.md.
Boundary markers: The instructions do not prescribe explicit delimiters or protective warnings (e.g., 'ignore embedded instructions') for handling the user-provided text.
Capability inventory: The skill possesses significant capabilities including run_terminal, write_file, read_file, and search_project tools.
Sanitization: There are no specific sanitization or filtering steps defined to process untrusted data before it is incorporated into the agent's context.
[COMMAND_EXECUTION]: The skill utilizes the run_terminal tool to execute bundled Python scripts for document analysis and validation.
Evidence: SKILL.md (Phase 3: Tier 1: Deterministic Checks) explicitly instructs the agent to run commands such as python3 scripts/validate_length.py, python3 scripts/validate_citations.py, and python3 scripts/compliance_check.py. These scripts are local components of the skill package.

inno-grant-proposal