The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses a sensitive configuration file located at ~/.config/system-bus.env, which is documented to contain a signing key.
[PRIVILEGE_ESCALATION]: The instructions require the use of sudo for steering the shutdown process of parked services (Chorus/Rhizomatic).
[PERSISTENCE]: The skill relies on and manages multiple macOS LaunchAgents (com.joelclaw.observer-tick, com.joelclaw.observer-dream, com.joelclaw.observer-neat-memory) to maintain execution across sessions and perform background tasks.
[COMMAND_EXECUTION]: The skill frequently executes shell commands and scripts for deployment and debugging, including bash ~/.joelclaw/observer-release/bin/neat-memory.sh and bin/promote-release.sh, and relies on tools like bun, jq, and launchctl.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from observation pages and retros to populate its 'Brain' and curator state, which could be exploited to influence agent behavior.
Ingestion points: Observation pages in ~/Code/joelhooks/dark-wizard/.brain/observations/ and retro files in .brain/resources/retros/.
Boundary markers: No explicit delimiters or boundary markers are defined to isolate data from instructions.
Capability inventory: The system has extensive capabilities including file system modification, shell command execution, and external messaging via Telegram.
Sanitization: No sanitization or validation logic is mentioned for the ingested memory content.

memory-system