The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process macro-level goals and requirements from the user, which constitutes untrusted external data.
Ingestion points: User requirements and responses during clarification rounds in SKILL.md.
Boundary markers: The protocol enforces the use of a Decision Ledger (tables with specific fields) and a Terminology First approach to delimit and define context.
Capability inventory: The skill instructs the agent to write to plan files and requirement documents. It includes a local Python script scripts/check-decision-ledger.py to validate the structure of these files.
Sanitization: The protocol requires explicit evidence (D-NN/E-NN/A-NN IDs) for all conclusions, reducing the likelihood of the agent blindly following instructions embedded within user requirements.
[COMMAND_EXECUTION]: The skill includes utility scripts and tests that execute shell commands.
Evidence: scripts/check-decision-ledger.sh and tests/test_clarification_protocol_contract.py utilize subprocess.run to execute the Python validator.
Context: These executions are confined to internal validation of the skill's own documentation and are triggered by the agent or developer during testing/verification phases rather than directly by external data flow. The implementation uses list-based arguments for subprocess.run, which prevents shell injection.

clarification-protocol