The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various CLI tools and local scripts to perform its core functions.
modules/deferred-capture.md invokes python3 scripts/deferred_capture.py to process course-of-action data.
modules/discussion-publishing.md utilizes the gh CLI (gh api graphql) to interact with GitHub Discussions.
modules/expert-roles.md employs tmux and asyncio.create_subprocess_exec to manage and spawn expert panel instances.
[DATA_EXFILTRATION]: Publishes deliberation summaries and session artifacts to an external platform.
modules/discussion-publishing.md transmits decision context, rationales, and expert contributions to GitHub Discussions. While this is a documented feature, it represents a channel for potentially sensitive organizational data to be sent to a remote service.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to malicious instructions embedded in the data it processes.
Ingestion points: Untrusted data enters via the user-provided decision topic and external files specified through the --files argument.
Boundary markers: The deliberation protocol and prompt templates lack explicit delimiters or instructions to ignore embedded commands within the analyzed content.
Capability inventory: The skill possesses capabilities for network operations (via gh), local command execution (python3, tmux), and file system writes (to the strategeion directory).
Sanitization: While the skill attempts to use safe subprocess execution, it does not explicitly sanitize the text content interpolated into expert prompts or discussion bodies.
[DYNAMIC_EXECUTION]: Performs runtime command resolution and execution.
modules/expert-roles.md dynamically determines CLI commands for the GLM expert (get_glm_command) based on the local environment and executes them using asyncio.create_subprocess_exec.

war-room