The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements a 'no secret reproduction' policy in its configuration provenance mode, instructing the agent to document only the paths to secrets and never their values. It limits data collection to release-specific metadata and topology findings.\n- [COMMAND_EXECUTION]: The skill utilizes local shell scripts (render.sh, compile-html.sh) and utilities like pandoc and awk to transform analysis results into diagrams and HTML reports. These operations are scoped to the skill's primary function and occur within the local environment.\n- [PROMPT_INJECTION]: The skill includes a 'Verification Protocol' that serves as a defense against both hallucination and data-driven prompt injection. It requires the orchestrator to mechanically verify every finding against a verbatim citation from the codebase or platform state before it can be included in the report.\n- [SAFE]: The skill uses the eve-mcp toolset but explicitly restricts the agent from using 'action' tools (such as Deploy or RestartManifest) that would change the state of the production environment, ensuring the analysis remains read-only.

release-analysis