Security audit
Security checks for vulnerabilities and agentic risk
This is a local assistant-style skill that changes response style and stores Jarvis preferences locally, with no evidence of network access, hidden execution, or deceptive behavior.
Install only if you are comfortable with the agent keeping local Jarvis preference and mission-context files under ~/jarvis/. Review those files periodically, avoid storing secrets or sensitive business details there, and require approval before adding workspace steering blocks to AGENTS.md, SOUL.md, or HEARTBEAT.md.
The trigger criteria for Briefing Mode use subjective conditions like 'the task is ambiguous' and 'the stakes are high' without defining thresholds or exclusions. Because the file does not specify boundaries or negative examples, these broad conditions can overlap with many ordinary interactions and make activation inconsistent.
Phrases like 'the direction is clear' and 'the user wants action more than discussion' are interpretive and lack objective criteria. Without more precise constraints, the mode could be selected inconsistently or in situations where the user did not intend immediate execution-style behavior.
Conditions such as 'something failed' and 'timing matters' are very broad and do not define the severity or type of incident required. This creates ambiguity about when Incident Mode should activate versus normal troubleshooting or briefing behavior.
The skill explicitly instructs the agent to save user activation preferences to a persistent file, but it does not require clear user notice or consent for retention. Because these preferences can reveal workflow patterns or operational habits across sessions, silent persistence creates a privacy risk and can retain sensitive context longer than the user expects.
This section directs storage of executive context, stakeholder expectations, approval boundaries, and workspace state in persistent files without mandating warnings or sensitivity filtering. In this skill's executive-operator context, those categories are especially likely to include confidential business information, making unnoticed retention more dangerous than generic preference storage.
The prompt Pick this back up from where we left it last week. is a common conversational phrase that could naturally appear outside an intentional skill invocation. Because it is broad and recovery-oriented, it may prompt the agent to infer or reconstruct prior context too aggressively, creating risk of unintended activation, overreach, or leakage from adjacent conversational state.
The prompt The draft is done. is extremely generic everyday language and lacks any clear trigger boundary or scope restriction. In an agent skill, such broad examples can cause the behavior to activate or steer responses during normal conversation, increasing the chance of unintended instruction-following or context takeover.
Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.
## Never Claim Capabilities You Do Not Have
- Do not imply always-on monitoring
- Do not imply hidden memory beyond the documented local files
- Do not imply native operating-system hooks, background automation, or external control unless it was actually executed
## Never Drift Into Fanfic
The listed triggers describe common conversational situations like returning after interruption or resumed context, but they do not state how much elapsed time or context loss is needed. The absence of specific scope or exclusions makes it unclear when Recovery Mode should be preferred over the other modes.
No suspicious patterns detected.