The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of fluentci execute to run task plans. According to the instructions in SKILL.md and references/native-plans.md, these tasks are executed directly on the host system with user-defined programs (e.g., /usr/bin/make), working directories, and environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill processes JSON-formatted task plans which define executable commands. This creates a surface for indirect injection if the plans are derived from untrusted sources.
Ingestion points: The agent interacts with plan.json files which define the task graph (documented in references/native-plans.md).
Boundary markers: The skill recommends using the --dry-run flag to preview task selection and logic before actual execution (documented in SKILL.md).
Capability inventory: The fluentci execute tool has the capability to launch arbitrary processes on the host machine and manage filesystem paths for run journals and logs (documented in SKILL.md).
Sanitization: The instructions advise the user to inspect the plan producer and verify SDK/engine capabilities, but there are no automated sanitization or validation steps described for the plan content itself.

fluentci-dag-caching