The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents several shell recipes using the roundfix CLI tool for repository inspection, plan generation, and configuration application. These are designed for interactive use, and the skill's disable-model-invocation setting prevents the AI from executing these commands autonomously.\n- [EXTERNAL_DOWNLOADS]: The skill references workflows for restoring 'Repository Skill Sets' and synchronizing 'canonical setup snapshots' from external sources. These operations are protected by a requirement for explicit maintainer authorization and the verification of Plan Digests before any changes are written.\n- [DATA_EXFILTRATION]: The instructions explicitly identify generated Baseline Plan files as sensitive artifacts that may contain repository policy. It implements a safety protocol by requiring manual approval and verification of these plans, ensuring that sensitive data is not mishandled or inadvertently applied.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes repository-authored rules and documentation.\n
Ingestion points: Processes instructions from 'Source Baseline Entry' rules and documents in docs/agents/.\n
Boundary markers: Includes warnings for nested instruction carriers and requires manual review for all rule classifications.\n
Capability inventory: Performs controlled file writes to the repository and utilizes the roundfix binary for processing.\n
Sanitization: Uses a strict 'Plan Digest' confirmation system where the human maintainer must verify the exact content of all changes before the apply command will execute.

setup-context-driven