The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of various shell commands and scripts located within the project's repository.
Specifically, it executes .guardrails/scripts/check-trace.sh and .guardrails/scripts/check-ids.sh.
It also executes arbitrary commands defined in .guardrails/config.yaml such as verify_commands and coverage_command.
[DYNAMIC_EXECUTION]: The agent is instructed to run commands dynamically determined by the configuration files within the repository. This is used to perform full test suites, type checks, and linters as part of the verification gate.
[INDIRECT_PROMPT_INJECTION]: The skill processes project-provided data such as command outputs, log files, and git status results.
Ingestion points: .guardrails/config.yaml, $TMPDIR/<branch>-gate.log, git status output, and subagent dispatch reports.
Boundary markers: The skill specifies a fixed shape for the gate summary and requires raw command output, but lacks explicit prompt delimiters for the ingested data.
Capability inventory: The skill has the capability to execute shell commands and scripts through subagents.
Sanitization: The skill mitigates risks by requiring specific metrics like pass, fail, and skip counts and raw output tails, rather than relying on exit codes or subjective claims.

verify-before-merge