The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands defined in repository-specific adapter files (adapter.env) using eval. This occurs in scripts/gate-build.sh, scripts/gate-static.sh, and scripts/token.sh. This behavior is central to the skill's purpose as a build and test orchestration tool.
[DYNAMIC_EXECUTION]: The skill instructions in SKILL.md direct the agent to analyze the target repository and author configuration files (adapter.env) that contain executable shell commands. The scripts then dynamically load and execute these commands to perform builds and tests.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from external API endpoints during Gate 4 (Scenario testing).
Ingestion points: API response bodies and headers are captured by scripts/call.sh and stored as receipts.
Boundary markers: While no specific prompt injection boundary markers are used, the skill implements a redaction library (scripts/lib.sh) to prevent sensitive data from being processed.
Capability inventory: The skill possesses the ability to execute shell commands (via the adapter), write files to the .verify/ directory, and perform network requests using curl.
Sanitization: The vf_redact function in scripts/lib.sh sanitizes credentials (JWTs, Authorization headers, cookies, and secret fields) using sed regex patterns before the data is saved or reported, reducing the risk of data exfiltration via prompt injection.

verify