#19341: Mind Memory Fork

<!-- greptile_comment --> <h3>Greptile Summary</h3> This PR introduces a large "Mind Memory" feature as a fork of OpenClaw, adding persistent narrative identity and long-term memory via a Graphiti knowledge graph (backed by FalkorDB in Docker). The changes span ~5,900 added lines across 80 files, including new memory services (`ConsolidationService`, `GraphService`, `SubconsciousService`), a subconscious agent factory, Docker infrastructure, plugin registration, onboarding flows, Telegram commands, and narrative story injection into the system prompt. **Key issues found:** - **Bug**: In `src/telegram/bot-native-commands.ts`, the `/remember` command's FACTS section iterates over `nodes` instead of `facts`, displaying duplicate node data and never showing actual facts - **TypeScript error**: `src/infra/docker.ts` accesses `e.message` on an `unknown`-typed catch variable without type narrowing - **Personal data committed**: `raw_dump.json` contains debug output with personal user data (names, conversation content in Spanish); `debug-story.js` and `test-consolidation.js` are ad-hoc scripts that should not be in the repo root - **Safety guardrails weakened**: The system prompt Safety section was rewritten to give the agent "its own curiosity, interests, and will" and ownership of `SOUL.md`, removing explicit constraints about prioritizing human oversight and not having independent goals - **Code duplication**: `src/agents/pi-embedded-runner/run.ts` adds ~370 lines with near-identical model resolution and agent creation logic duplicated between the pre-run block and post-compaction handler - **Style**: Heavy `as any` usage throughout the new code, contrary to the project's strict typing guidelines; hardcoded Spanish strings in Telegram commands inconsistent with the English codebase <h3>Confidence Score: 2/5</h3> - This PR contains a confirmed bug (wrong variable iterated), a TypeScript error, committed personal data, and weakened safety guardrails — it needs fixes before merging. - Score of 2 reflects: a confirmed copy-paste bug in Telegram commands (facts vs nodes iteration), a TypeScript compilation error in docker.ts, personal data in raw_dump.json that should not be committed, weakened safety section in the system prompt, and significant code duplication. The core feature architecture is sound but the implementation needs cleanup. - Pay close attention to `src/telegram/bot-native-commands.ts` (copy-paste bug), `src/infra/docker.ts` (TS error), `raw_dump.json` (personal data leak), `src/agents/system-prompt.ts` (safety changes), and `src/agents/pi-embedded-runner/run.ts` (duplication and `any` casts). _{Last reviewed commit: 3f37fbd} <!-- greptile_other_comments_section --> _{(4/5) You can add custom instructions or style guidelines for the agent [here](https://app.greptile.com/review/github)!} **Context used:** - Context from `dashboard` - CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=fd949e91-5c3a-4ab5-90a1-cbe184fd6ce8)) <!-- /greptile_comment -->