night-autopilot
Warn
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: MEDIUMPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PERSISTENCE]: The skill implements automated task scheduling by instructing the agent to configure system persistence mechanisms, including
cronjobs,launchdagents, and platform-specific timers (CronCreate/CronUpdate) as described inSKILL.md. This allows the agent to operate autonomously without an active user session.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external, potentially untrusted sources such as project issue trackers and local task files mentioned inSKILL.md.\n - Ingestion points: GitHub issues, project-specific
TASKSfiles, and conversation history.\n - Boundary markers: The skill requires the definition of a "Safety boundary" and "Forbidden list" in
SKILL.mdbefore operation.\n - Capability inventory: Execution of Git commands, GitHub API interactions (
gh api), file system modifications (state anchor updates), and management of subagent workers inSKILL.md.\n - Sanitization: Employs domain-specific audit criteria, such as automated test suites for code and mandatory human review gates for legal documents, though it does not explicitly mention technical sanitization of the input data itself.\n- [COMMAND_EXECUTION]: The agent uses various command-line tools (
git,gh,claude,codex) and system utilities to manage project states and automate the scheduling of its own execution loop as outlined inSKILL.md.
Audit Metadata