The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions explicitly mandate the execution of bash commands for hash calculation and drift detection logic.
Evidence: Section 9 'Hash Calculation' commands the agent to execute sha256sum <file_path> | cut -d' ' -f1 and yq '.<section_name>' <file_path> | sha256sum | cut -d' ' -f1.
Evidence: Direct instruction 'CRITICAL: Execute actual bash commands. DO NOT write placeholder values.' forces the agent into a shell execution context.
[COMMAND_EXECUTION]: Potential for command injection if file paths or section names extracted from TSPEC documents are not sanitized before being passed to shell utilities like yq or sha256sum.
[DATA_EXFILTRATION]: The skill performs local file reads and writes (creating reports and a .drift_cache.json file) but does not contain instructions for external network communication.
[PROMPT_INJECTION]: The skill processes untrusted external documents (TSPEC and SPEC files) that are used as inputs for review logic and hash calculation, creating a surface for indirect prompt injection.
Ingestion points: TSPEC markdown files in docs/10_TSPEC/ and SPEC yaml files in docs/09_SPEC/.
Boundary markers: None specified in the instructions.
Capability inventory: Shell command execution (sha256sum, yq, grep, cut), file write access for reports and cache files.
Sanitization: No evidence of validation or sanitization of document content before processing or shell interpolation.

doc-tspec-reviewer