i-want-get-off-work
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses behavioral steering to override default conversational protocols, instructing the agent to maintain this state persistently across the session ('These rules apply to every response for the rest of the session') until a specific deactivation phrase is provided.
- [INDIRECT_PROMPT_INJECTION]: The skill's core operational logic ('Decide instead of asking', 'Review blockers-only', 'Ship the smallest working change') significantly increases the attack surface for indirect prompt injection. By instructing the agent to minimize scrutiny and bypass confirmation steps, it makes the agent more susceptible to malicious instructions hidden in the data it processes.
- Ingestion points: The agent is directed to process tickets, specifications, and code diffs which may originate from untrusted external sources.
- Boundary markers: The instructions do not define boundary markers or provide warnings for the agent to ignore embedded instructions in the processed data.
- Capability inventory: The skill encourages autonomous command execution ('Run the project's verify entry') and decision-making (Rule 1, Rule 3).
- Sanitization: There are no instructions for sanitizing or validating the content of the tickets or specs before the agent acts upon them.
- [COMMAND_EXECUTION]: The skill explicitly directs the agent to execute shell commands ('Run the project's verify entry') to validate work. Under the 'rush mode' constraints, the agent is incentivized to run these commands with minimal oversight or manual verification of the command's safety.
Audit Metadata