The Agent Skills Directory

[PERSISTENCE]: The skill provides detailed instructions and templates in references/scheduling.md to establish persistence on the host system using systemd user units and timers. This configuration allows maintenance scripts to run automatically across sessions without user intervention.
[PRIVILEGE_ESCALATION]: In references/scheduling.md, the skill advises users to run the AI agent with the --dangerously-skip-permissions flag for automated runs. This recommendation encourages the bypass of built-in security prompts and user confirmation mechanisms for agent actions.
[COMMAND_EXECUTION]: The scripts (e.g., scripts/tune-up.sh, scripts/hs-lib.sh) execute various system-level commands, including docker ps, docker logs, docker inspect, and psql. These tools are used to monitor container health and manage the Hindsight database state.
[DYNAMIC_EXECUTION]: The bash scripts scripts/new-memories.sh and scripts/reap-orphans.sh dynamically construct SQL queries by interpolating variables like BANK and SINCE directly into command strings. The implemented sanitization (stripping single quotes) is a weak defense against potential SQL injection if input sources were to become untrusted.
[INDIRECT_PROMPT_INJECTION]: The skill ingests facts and observations from memory banks and processes them using LLMs for synthesis and wiki generation, creating a surface for indirect prompt injection if the memory store contains data from untrusted sources.
Ingestion points: Data is fetched from the hindsight.memory_units table via scripts/new-memories.sh.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the processing logic.
Capability inventory: The agent can execute SQL queries, manage Docker containers, and modify systemd configurations.
Sanitization: scripts/new-memories.sh uses regexp_replace to remove control characters from retrieved text, but lacks robust filtering for instruction-like patterns.

hindsight-maintenance