The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external software repositories and issue trackers, which could potentially contain malicious instructions intended to manipulate the agent's behavior.
Ingestion points: Untrusted source code, README files, and issue descriptions are read into the agent context, specifically when diagnosing fit or planning coding tasks (SKILL.md, references/run-and-delivery.md).
Boundary markers: The skill instructions mandate the creation of a 'run contract' with frozen acceptance criteria and non-goals. It also uses explicit --allow-* flags in its helper scripts to ensure the agent does not proceed without a recorded human approval for sensitive configurations like current-checkout mutation or auto-merging.
Capability inventory: The skill interacts with the zeroshot CLI, which has the capability to execute subprocesses (provider CLIs), modify the local filesystem via worktrees, and perform network operations like pushing Git branches and creating Pull Requests via forge CLIs.
Sanitization: The doctor mode in scripts/zeroshot.sh checks for the presence of credentials in environment variables without disclosing their values, and the trace_summary.py script facilitates auditing trace logs without exposing the actual message content or prompts to the agent.
[COMMAND_EXECUTION]: The skill provides automated assistance for constructing and executing zeroshot CLI commands while maintaining strict control over execution.
Evidence: The scripts/zeroshot.sh helper includes a preflight command that performs validation checks and outputs a shell-quoted command proposal. The instructions strictly forbid the agent from executing these commands automatically, requiring the user to review and approve the exact final command string.
[EXTERNAL_DOWNLOADS]: The skill guides the installation of the ZeroShot toolchain from the @the-open-engine organization.
Evidence: The installation procedures in references/product-and-installation.md specify pinned versions (e.g., @the-open-engine/zeroshot@6.45.0) to prevent version drifting and ensure the agent uses audited releases from the official registry.

zeroshot