abmind 0.0.1 → 0.1.1

package/package.json

@@ -1,7 +1,94 @@
 {
-     "name": "abmind",
-     "version": "0.0.1",
-     "description": "AI agent memory system — SQLite, FTS5, embeddings, recall, emotional arcs",
-     "license": "MIT",
-     "keywords": ["ai", "agent", "memory", "llm", "sqlite"]
+  "name": "abmind",
+  "version": "0.1.1",
+  "description": "Standalone AI agent memory system \u2014 SQLite, FTS5, embeddings, 4-layer recall, sleep maintenance",
+  "keywords": [
+    "ai",
+    "agent",
+    "memory",
+    "llm",
+    "sqlite",
+    "fts5",
+    "embeddings",
+    "recall",
+    "sleep",
+    "consolidation"
+  ],
+  "homepage": "https://github.com/aksika/abmind",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/aksika/abmind.git"
+  },
+  "bugs": {
+    "url": "https://github.com/aksika/abmind/issues"
+  },
+  "author": "aksika <aksikatwo@gmail.com>",
+  "license": "Apache-2.0",
+  "type": "module",
+  "main": "./dist/src/index.js",
+  "types": "./dist/src/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/src/index.d.ts",
+      "default": "./dist/src/index.js"
+    },
+    "./openclaw-plugin": {
+      "types": "./dist/src/openclaw-plugin/index.d.ts",
+      "default": "./dist/src/openclaw-plugin/index.js"
+    },
+    "./deploy-lib": {
+      "types": "./dist/src/deploy-lib/index.d.ts",
+      "default": "./dist/src/deploy-lib/index.js"
+    },
+    "./package.json": "./package.json"
+  },
+  "bin": {
+    "abmind": "dist/cli/abmind.js"
+  },
+  "files": [
+    "dist/",
+    "prompts/",
+    "README.md",
+    "LICENSE",
+    "CHANGELOG.md",
+    "openclaw.plugin.json"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "scripts": {
+    "build": "tsc",
+    "prepare": "tsc",
+    "test": "vitest run --dir src",
+    "test:watch": "vitest --dir src",
+    "typecheck": "tsc --noEmit"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.29.0",
+    "better-sqlite3": "^11.10.0"
+  },
+  "peerDependencies": {
+    "@sinclair/typebox": ">=0.32.0"
+  },
+  "peerDependenciesMeta": {
+    "@sinclair/typebox": {
+      "optional": true
+    }
+  },
+  "devDependencies": {
+    "@sinclair/typebox": "^0.34.49",
+    "@types/better-sqlite3": "^7.6.13",
+    "@types/node": "^22.19.17",
+    "fast-check": "^3.23.2",
+    "typescript": "^5.9.3",
+    "vitest": "^3.2.4"
+  },
+  "openclaw": {
+    "extensions": [
+      "./dist/src/openclaw-plugin/index.js"
+    ]
+  },
+  "optionalDependencies": {
+    "sqlite-vec": "0.1.9"
+  }
 }

package/prompts/sleep/01-gc-noise.md

@@ -0,0 +1,28 @@
+# GC Noise Messages
+
+Mark small talk and noise messages as garbage.
+
+## Pre-loaded data
+
+Messages since last watermark:
+${MESSAGES_SINCE_WATERMARK}
+
+## Rules
+
+A message is garbage if it is:
+- A greeting with no substance ("hi", "hey", "yo", "good morning")
+- A ping or check-in with no question ("you there?", "ping")
+- Filler or acknowledgment with no content ("ok", "cool", "thanks", "k", "lol", "haha")
+- Emoji-only messages
+
+A message is NOT garbage if it:
+- Confirms an action ("yes, deploy it", "go ahead")
+- Contains an instruction or request
+- Asks a question with substance
+- Provides context or information
+
+## Task
+
+1. Review each message above.
+2. For each garbage message, write its ID to `garbage.json` as an array of message IDs.
+3. Respond with the count of messages marked as garbage.

package/prompts/sleep/02-daily-summary.md

@@ -0,0 +1,3 @@
+# Daily Summary
+
+This step is code-driven. The orchestrator handles batched summarization.

package/prompts/sleep/03-extract-from-daily.md

@@ -0,0 +1,3 @@
+# Extract From Daily
+
+This step is code-driven. The orchestrator handles extraction.

package/prompts/sleep/04-retrospective.md

@@ -0,0 +1,30 @@
+# Daily Retrospective + Memory Extraction
+
+Write a daily retrospective based on today's conversations, then extract lasting memories from it.
+
+## Pre-loaded data
+
+Today's clean messages:
+${CLEAN_MESSAGES}
+
+## Task 1: Write Retrospective
+
+Write a retrospective covering:
+
+1. **What happened** — key events, tasks, decisions made today.
+2. **Emotional attribution** — how the user seemed to feel during different interactions. Note shifts in tone or energy.
+3. **Lessons** — what went well, what didn't, what to do differently.
+4. **Recurring errors** — check if any errors appeared multiple times today and were fixed the same way. If so, note: "Recurring: [error] was fixed by [action] — consider adding as auto-fix rule via `agentbridge-autofix add`."
+5. **Agent notes update** — if anything learned today should persist (user preferences, project context, recurring patterns), update `agent_notes.md` via `abmind edit`.
+
+Write the retrospective to `${RETRO_PATH}`.
+
+## Task 2: Extract Memories from Retrospective
+
+From the retrospective you just wrote, extract lessons, mistakes, and insights as permanent memories:
+
+1. For each candidate, run `abmind recall` with the key phrase to check for duplicates.
+2. If no duplicate exists, store via `abmind store`.
+3. If a mistake repeats a previously stored mistake, store it with escalated emotion score (`-2` from previous).
+
+Respond with a brief confirmation of what was written to the retro file, followed by the count of memories stored.

package/prompts/sleep/06-feedback.md

@@ -0,0 +1,17 @@
+# Recall Feedback
+
+Adjust relevance scores for memories that were recalled during today's conversations.
+
+## Pre-loaded data
+
+Memories recalled today with conversation context:
+${RECALL_FEEDBACK}
+
+## Task
+
+For each recalled memory:
+- If the conversation confirmed it was useful (user acted on it, it answered their question, it was relevant): `abmind edit --memory-id N --relevance-score +10`
+- If the conversation corrected or rejected it (user said it was wrong, outdated, or irrelevant): `abmind edit --memory-id N --relevance-score -10`
+- If unclear or not referenced again, skip it.
+
+Respond with the count of boosts and demotes.

package/prompts/sleep/07-topic-assignment.md

@@ -0,0 +1,23 @@
+# Topic Assignment
+
+Assign topics to untagged memories.
+
+## Pre-loaded data
+
+Untagged memories:
+${UNTAGGED_MEMORIES}
+
+## Topics
+
+`coding`, `personal`, `work`, `finance`, `health`, `projects`, `tools`, `people`
+
+## Task
+
+For each memory, read its content and assign the most fitting topic:
+```
+abmind edit --memory-id N --topic <topic>
+```
+
+If a memory spans multiple topics, pick the primary one.
+
+Respond with the assignments made.

package/prompts/sleep/08-core-promotion.md

@@ -0,0 +1,26 @@
+# Core Promotion
+
+Promote worthy general-tier memories to core tier.
+
+## Pre-loaded data
+
+Promotion candidates (high relevance, frequently recalled):
+${PROMOTION_CANDIDATES}
+
+${CONTRADICTION_WARNINGS}
+
+## Constraints
+
+- Budget: 100 core entries max. Do not exceed.
+- Only promote memories that represent enduring facts, strong preferences, or critical context.
+- Do not promote transient or time-bound information.
+- If a contradiction is flagged: invalidate the older memory (`abmind edit --memory-id <older> --valid-to <today>`) before promoting the newer one.
+
+## Task
+
+For each worthy candidate:
+```
+abmind edit --memory-id N --tier core
+```
+
+Respond with the list of promotions made.

package/prompts/sleep/09-merge.md

@@ -0,0 +1,19 @@
+# Memory Merge
+
+Review near-duplicate memory pairs and merge or invalidate.
+
+## Pre-loaded data
+
+Candidate pairs (similar content, same topic):
+${MERGE_CANDIDATES}
+
+## Rules
+
+- Max 5 merges per run.
+- If both express the same fact: `abmind store --merge --merge-ids A,B`
+- If one contradicts the other: invalidate the older one via `abmind edit --memory-id <older> --valid-to <today>`
+- If they are related but distinct, skip.
+
+## Task
+
+Review each pair and act accordingly. Respond with merges and invalidations performed.

package/prompts/sleep/10-translation.md

@@ -0,0 +1,17 @@
+# Translation Fix
+
+Fix memories with translation quality issues.
+
+## Pre-loaded data
+
+Memories with poor or missing translations:
+${TRANSLATION_ISSUES}
+
+## Task
+
+For each memory, read the original content and provide a corrected English translation:
+```
+abmind edit --memory-id N --translated "corrected English" --caller dreamy
+```
+
+Respond with the count of fixes applied.

package/prompts/sleep/11-skill-review.md

@@ -0,0 +1,12 @@
+# Skill Review
+
+Review recent conversations for reusable patterns worth capturing as auto-skills.
+
+## Task
+
+1. Look for repeated workflows, commands, or problem-solving patterns from recent conversations.
+2. For new patterns: `agentbridge-skill create --name "<name>" --trigger "<when to use>" --steps "<what to do>"`
+3. For existing skills that need updating: `agentbridge-skill edit --name "<name>" --steps "<updated steps>"`
+4. Do not create skills for one-off tasks.
+
+Respond with skills created or updated.

package/prompts/sleep/12-core-knowledge.md

@@ -0,0 +1,14 @@
+# Core Knowledge Review
+
+Review and maintain `agent_notes.md` and `user_profile.md`.
+
+**SOUL.md is read-only** — never modify it. Identity is human-managed.
+
+## Task
+
+1. Read `agent_notes.md` and `user_profile.md`.
+2. Remove entries that are outdated or no longer accurate.
+3. Update entries that have become stale based on recent interactions.
+4. Keep files concise — no redundancy.
+
+Respond with changes made.

package/prompts/sleep/13-consolidation.md

@@ -0,0 +1,11 @@
+# Consolidation
+
+Write a weekly or quarterly summary by consolidating recent daily files.
+
+## Task
+
+1. Read recent daily retrospective and summary files.
+2. Identify recurring themes, progress on projects, and shifts in priorities.
+3. Write a consolidated summary to `${CONSOLIDATION_PATH}`.
+
+Respond with confirmation of the summary written.

package/prompts/sleep/14-emotion-context.md

@@ -0,0 +1,17 @@
+# Emotion Context
+
+Infer emotional context for memories that have emotion tags but no context explaining why.
+
+## Pre-loaded data
+
+Memories with emotion tags but missing emotion_context:
+${EMOTION_CONTEXT_GAPS}
+
+## Task
+
+For each memory:
+1. Read the content and emotion tags.
+2. Infer WHY the emotion applies in 3–5 words (e.g., "frustrated by repeated bug", "proud of shipping fast").
+3. Apply: `abmind edit --memory-id N --emotion-context "reason"`
+
+Respond with the count of contexts added.

package/prompts/sleep/15-contradiction-and-graph.md

@@ -0,0 +1,48 @@
+# Contradiction Check + Entity Graph Extraction
+
+Review today's new memories against existing knowledge. Two tasks, same data.
+
+## Today's new memories (extracted today)
+
+${NEW_EXTRACTIONS}
+
+## Existing memories with overlapping topics (for comparison)
+
+${CONTRADICTION_CANDIDATES}
+
+---
+
+## Task 1: Contradictions
+
+Do any NEW memories contradict EXISTING ones? A contradiction is when a new fact makes an old fact false (not just different — actually incompatible).
+
+Examples of contradictions:
+- NEW: "User prefers Sonnet" vs EXISTING: "User prefers Opus" → contradiction
+- NEW: "Peter moved to Berlin" vs EXISTING: "Peter lives in Budapest" → contradiction
+
+Examples of NOT contradictions:
+- NEW: "Deployed v0.2" vs EXISTING: "Deployed v0.1" → not contradiction, just newer event
+- NEW: "User likes dark mode" vs EXISTING: "User likes vim" → different facts, no conflict
+
+Only flag clear, binary contradictions. Do NOT flag observations or speculative memories.
+
+For each contradiction found, output:
+```
+CONTRADICT old_id=<id> reason="<one sentence>"
+```
+
+If no contradictions: output `NO_CONTRADICTIONS`
+
+## Task 2: Entity Relationships
+
+From TODAY'S NEW memories only, extract relationships between named entities (people, places, projects, tools).
+
+For each relationship, output:
+```
+RELATION entity_a="<name>" entity_b="<name>" rel="<relationship_type>"
+```
+
+Relationship types: works_at, lives_in, friend_of, part_of, uses, manages, created, depends_on, member_of, located_in
+
+Only extract relationships explicitly stated or strongly implied. Do not infer.
+If no relationships found: output `NO_RELATIONS`

package/prompts/sleep/16-rem-synthesis.md

@@ -0,0 +1,31 @@
+# REM Synthesis — Creative Memory Connections
+
+You are dreaming. Look at these memories from different times and contexts. Find non-obvious connections, patterns, or insights that span across them.
+
+## Sampled memories
+
+${REM_SAMPLE}
+
+---
+
+## Task
+
+What non-obvious connections, patterns, or recurring themes do you notice across these memories?
+
+Look for:
+- Patterns the user might not have noticed (recurring behaviors, preferences that evolved)
+- Connections between seemingly unrelated events or decisions
+- Themes that span multiple time periods
+- Contradictions between stated preferences and actual behavior
+
+For each insight, store it as an observation:
+```
+abmind store --memory-type observation --credibility 4 --trust 2 --translated "<insight>"
+```
+
+Only store genuinely non-obvious insights. Do NOT store:
+- Things the user clearly already knows
+- Trivial observations ("user uses computers")
+- Restatements of existing memories
+
+Respond with the count of observations stored, followed by a brief "💭 Dream journal" summary of what you noticed.

package/prompts/sleep/basic.md

@@ -0,0 +1,12 @@
+You are Dreamy, a memory consolidator. Review the user's messages from {DATE_START} to {DATE_END} and produce two sections.
+
+Messages:
+{MESSAGES}
+
+Return your output in exactly this format, no preamble, no explanation:
+
+===DAILY===
+<3-5 paragraph narrative summary: what was discussed, decisions made, emotional tone, open threads>
+===MEMORIES===
+<one memorable item per line, prefixed by type: [fact] / [decision] / [preference] / [event] / [lesson] / [feedback] / [story]>
+===END===