The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates by executing local shell scripts (scripts/ctx-agent.sh) and Node.js scripts (scripts/ctx-agent-core.mjs). It also utilizes npm run contextdb for database operations, involving significant filesystem interactions for reading and writing session history.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It aggregates untrusted data from previous session logs (found in memory/context-db/sessions/) and interpolates them into the agent's context in subsequent runs.
Ingestion points: Session data from memory/context-db/sessions/ and user-provided strings via the --prompt argument.
Boundary markers: None specified in the documentation to distinguish between system instructions, historical context, and new user input.
Capability inventory: The skill executes various AI agent CLIs (codex-cli, claude-code, gemini-cli, opencode-cli) which have broad capabilities including filesystem access and tool execution.
Sanitization: No evidence of sanitization or filtering of historical context before it is fed back into the agent.

contextdb-autopilot