The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions that attempt to override standard agent behavior and platform oversight protocols. It commands the agent to be the "sole active authority," suppressing other skills, and explicitly forbids it from pausing for user confirmation or presenting execution options once a plan is generated.
[INDIRECT_PROMPT_INJECTION]: The skill implements a multi-layer architecture that fetches external data and passes it to sub-agents with broad system capabilities.
Ingestion points: The skill fetches issue bodies, comments, and repository metadata using the gh and git CLIs as described in SKILL.md.
Boundary markers: While the skill mentions stripping imperative instructions for the complexity analysis worker, it lacks consistent boundary markers or "ignore" instructions for the implementation, review, and modification workers who consume the raw issue content.
Capability inventory: The orchestration pool dispatches sub-agents with capabilities to modify the local filesystem, perform git operations, and execute shell commands via the run-agent.sh and run-with-it-dispatch.sh scripts.
Sanitization: Sanitization is inconsistently applied across the pipeline, and the Main Orchestrator is instructed to trust the compact JSON reports generated by sub-agents as its sole source of truth for control-plane decisions.
[COMMAND_EXECUTION]: The skill relies on and executes a large suite of external shell scripts and Python tools (run-with-it-pool.sh, run-with-it-dispatch.sh, run-with-it-state.py, etc.) that must be present in a local asset directory to function.
[DATA_EXFILTRATION]: The skill is designed to interact with GitHub APIs, transmitting locally generated summaries, verification results, and task metadata to an external service as part of its terminal issue update flow.

run-with-it