The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses sensitive local credential storage to authenticate with repository hosts.
The script scripts/patrol-snapshot.mjs reads authentication tokens and configurations from ~/.agents/env/forgejo.sh.
The script scripts/patrol-snapshot.mjs utilizes git credential fill to programmatically retrieve stored passwords for Forgejo and other repository instances.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from version control systems, creating a vulnerability surface for indirect prompt injection.
Ingestion points: Aggregates thread history via t3_thread_read, pull request reviews and comments via gh pr view, and Forgejo/Gitea API responses.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when incorporating external text into the agent's context or when generating messages in references/messages.md.
Capability inventory: The agent possesses high-privilege capabilities including sending messages (t3_thread_send), launching new threads (t3_thread_launch), modifying model configurations (t3_thread_configure), and archiving/settling threads (t3_thread_organize).
Sanitization: While content is clipped for length, no structural sanitization or escaping of untrusted input is performed before interpolation into prompts.
[COMMAND_EXECUTION]: The skill executes several system and version control command-line utilities to maintain state awareness.
Uses hostname and sysctl for device identification.
Uses git, gh (GitHub CLI), and fj (Forgejo CLI) to query repository metadata and PR health.
Uses df to monitor disk space before performing handoff operations.
[DYNAMIC_EXECUTION]: The skill relies on local Node.js scripts to perform its core logic, which involves complex data aggregation and runtime model switching.
Executes scripts/patrol-snapshot.mjs to collect thread and PR snapshots.
Executes scripts/patrol-tier.mjs to perform model tier switching and thread recovery actions.

t3-patrol