The Agent Skills Directory

[SAFE]: The skill maintains a local state using a tracking file (data/tcm-constitution-tracker.json) and is restricted to 'Read' and 'Write' operations, limiting the impact of any potential data manipulation and ensuring all data remains within the local environment.
[SAFE]: No network-based exfiltration or external content downloads were identified. The skill does not communicate with any remote servers or fetch external code.
[SAFE]: The instructions include clear medical disclaimers and constraints on the agent's behavior (e.g., "Do not diagnose TCM diseases", "Do not prescribe herbal medicine"), which helps prevent the agent from providing unsafe medical advice or exceeding its intended scope.
[SAFE]: No obfuscation, hidden URLs, or malicious prompt injection patterns were found in the documentation, examples, or schema files.
[NO_CODE]: The skill consists entirely of markdown instructions and JSON data structures without any executable scripts or binary dependencies, which minimizes the technical risk of remote code execution or privilege escalation.
[SAFE]: The skill has a surface for indirect prompt injection (Category 8) as it processes user answers and stores them in a JSON file for later analysis, however this is mitigated by the following factors:
Ingestion points: User-provided answers during the 'assess' flow and subsequent reads from data/tcm-constitution-tracker.json.
Boundary markers: The skill uses a defined JSON schema (schema.json) to structure and validate all stored data.
Capability inventory: The skill is restricted to 'Read' and 'Write' tools; no shell execution, network access, or administrative capabilities exist.
Sanitization: Data is constrained by the strict JSON schema, which reduces the risk of arbitrary instruction injection into the agent's context.

tcm-constitution