The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines several shell helper functions (gitea_auto, gitea_use, gitea) that execute system commands including curl, git, jq, grep, and sed. These are used to discover Gitea instances from repository remotes, manage authentication tokens, and interact with the REST API.
[DATA_EXFILTRATION]: The skill transmits sensitive Personal Access Tokens (PAT) to Gitea instances defined by the user in environment variables or detected from git remotes. While this is the intended functionality, it involves sending credentials to arbitrary external hosts. The skill mitigates risk by instructing the agent to never include tokens in query strings and to use secure headers.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from Gitea instances, including issue titles, comments, pull request descriptions, and repository file contents.
Ingestion points: Data is fetched via GET requests to various API endpoints (e.g., /repos/{owner}/{repo}/issues, /repos/{owner}/{repo}/contents/{path}) and processed using jq before being presented to the agent.
Boundary markers: No explicit boundary markers or instructions to disregard embedded commands in the fetched data are present in the skill's prompts.
Capability inventory: The agent has the capability to perform network operations (curl) and local file operations (git) which could be abused if malicious instructions in the data are followed.
Sanitization: The skill relies on jq for parsing structured JSON data but does not implement specific sanitization or filtering of the natural language content within that data.

gitea