The Agent Skills Directory

[NO_CODE]: The skill bundle is composed exclusively of Markdown documentation, YAML/JSON configuration files, and templates. It does not ship with any Python scripts, JavaScript, binaries, or other executable code.
[SAFE]: No network operations, external API calls, or remote resource downloads are present. The skill relies on local relative paths for all routing and operates within the bounds of the host's local file system as stated in the requirements.
[SAFE]: There are no hardcoded credentials, secrets, or sensitive configuration files. The framework proactively instructs the agent to treat artifacts as private and establishes an 'External-sharing gate' for sanitizing data before sharing (references/communication.md).
[PROMPT_INJECTION]: Evaluated the surface for indirect prompt injection given the skill's ingestion of external stakeholder data.
Ingestion points: Untrusted stakeholder feedback and workflow observations are intended to be recorded in files within the templates/ directory.
Boundary markers: While the templates do not use technical delimiters, the 'Epistemic and communication contract' (SKILL.md) enforces strict labeling of facts versus inferences.
Capability inventory: The skill routes to powerful specialists such as remote-systems-administration and secure-software-engineering (manifest.yaml).
Sanitization: Risk is mitigated by the 'Authority Table' (references/authority-and-escalation.md), which requires explicit, documented approval for any action involving security boundaries, privacy, or system access, and the mandatory sanitization review process.

forward-deployed-engineering