The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for workflows that ingest and process output from external AI models (e.g., via the claudish CLI). This ingestion of untrusted data creates a potential surface for indirect prompt injection if the model output contains malicious instructions that influence subsequent automated steps.
Ingestion points: Output from external models captured via the claudish tool as documented in SKILL.md.
Boundary markers: The provided implementation patterns do not include specific delimiters or "ignore embedded instructions" warnings for the ingested content.
Capability inventory: The skill logic includes file system writes (ai-docs/grok-review.md, ai-docs/partial-review.md), shell command execution via timeout and bash, and process management (handling SIGINT).
Sanitization: No specific sanitization, validation, or filtering of the model-generated content is implemented in the provided examples before the data is processed or saved.
[EXTERNAL_DOWNLOADS]: The documentation references the installation of the claudish CLI tool via npm install -g claudish to enable multi-model functionality. It also describes querying the OpenRouter API (openrouter.ai) for credit balance checks. These are standard external resources for the documented workflow.

error-recovery