The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill uses sudo -n within scripts/sweep.sh to perform privileged operations, such as reading the kernel ring buffer (dmesg) and auditing the root user's crontab.
[DATA_EXPOSURE]: The skill accesses sensitive system telemetry, including kernel logs, environment variables, and comprehensive process lists. It also explicitly references secret storage mechanisms, including local secrets.zsh files and 1Password vaults, as sources for resolving environment variables.
[COMMAND_EXECUTION]: The skill relies on extensive execution of system administration tools such as systemctl, docker, and srvls to manage services and modify system state, including destructive actions like pruning containers and volumes.
[DATA_EXFILTRATION]: The skill transmits diagnostic metadata and 'memories' of the maintenance process to the vendor's external Hindsight API (api.hs.delo.sh).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted system data from journalctl and docker logs to drive its forensic and diagnostic phases. It lacks sanitization or boundary markers for this data, while maintaining the authority to perform destructive system actions based on its interpretation of those logs.
Ingestion points: journalctl output, docker logs, and the generated sweep report file in /tmp/.
Boundary markers: None identified in the instructions or scripts.
Capability inventory: High-privilege command execution (sudo, systemctl, docker), file modification in references/, and the ability to stop or remove containers.
Sanitization: None identified; log output is parsed and acted upon directly by the agent.

server-maintenance