The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script scripts/check-output.py designed to validate the presence of specific sections in generated markdown files. The script is executed via command line and uses standard file I/O operations (pathlib.read_text()). It does not perform any network operations or sensitive file access.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest customer-related data (usage signals, health scores, and incident reports) to generate playbooks. While it processes untrusted input, it lacks capabilities for high-risk operations like network exfiltration or shell execution of that data. The threat is mitigated by the analytical nature of the output (markdown templates) and the lack of automated tool triggers based on the processed data.
Ingestion points: Data enters the context through user-provided descriptions of their customer success function, health metrics, and crisis scenarios in the primary agent dialogue.
Boundary markers: The skill uses a templates/output-template.md to structure the agent's output, providing clear sections that separate the agent's reasoning from the user's data.
Capability inventory: The skill only includes a local file validator (scripts/check-output.py). There are no capabilities for writing to arbitrary file paths, making network requests, or spawning subprocesses based on user input.
Sanitization: None detected; the skill relies on the underlying LLM's safety guardrails for text processing.

cs-playbooks