The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from code diffs and environment configurations during its reconciliation phase, which could be manipulated to influence agent behavior.
Ingestion points: references/configuration-preflight.md instructs the agent to extract configuration keys and secret names from source code files, .env.example files, and CI workflow definitions.
Boundary markers: There are no specific delimiters or instruction-ignore warnings defined for the data extracted from the code diffs.
Capability inventory: The skill can execute powerful infrastructure commands via gh and fly CLIs, and runs arbitrary smoke test and rollback commands defined in the release process.
Sanitization: The skill includes strong instructions to handle "names only, never values" and to use non-secret canary placeholders, which mitigates the risk of direct credential exposure.
[COMMAND_EXECUTION]: The skill facilitates the execution of sensitive commands related to secret management and deployment validation.
Evidence: references/configuration-preflight.md explicitly lists commands such as gh secret list, gh secret set, and fly secrets set to be used by the agent to verify and provision environment secrets.

release-gate