Explain the installed plugin

Read ${CLAUDE_PLUGIN_ROOT}/README.md, especially “What this plugin runs, stores and sends”, “Automatic capture controls”, and “Logs and state”. Those shipped instructions describe this installed version. Run "${CLAUDE_PLUGIN_ROOT}/scripts/cognee-doctor.sh" --json when the user asks about their connection or settings. Do not expose keys or raw private memory in an answer.

Explain the paths relevant to the question:

• Prompts, answers, and selected tool traces are captured by hooks, with credential redaction and sensitive-file exclusions. COGNEE_CAPTURE=false opts out.
• Cloud calls go to COGNEE_BASE_URL using COGNEE_API_KEY; local mode uses a local server. Cloud credit status comes from the Cognee platform API.
• Recall injects selected memory into Claude's context. The injection budget drops whole entries and the session ledger avoids repeating cloud memory until context is cleared or compacted.
• Local logs and markers live in ~/.cognee-plugin/claude-code/. Server memory stays until explicitly deleted. cognee-report measures observed outcomes.
• Cloud setup manages native auto-memory only when there is no explicit user choice; COGNEE_DISABLE_AUTO_MEMORY=false restores the plugin-owned setting.

Distinguish an accepted capture, a landed graph improve, and recalled context. Consult the installed README for background retries/cache behavior before claiming it is available. Explain limitations directly when telemetry cannot establish an outcome.