claude-memory-agent 2.0.0

package/agent_card.py

@@ -0,0 +1,512 @@
+"""A2A Agent Card definition for Claude Memory Agent."""
+
+AGENT_CARD = {
+    "name": "Claude Memory Agent",
+    "description": "Persistent semantic memory for Claude Code sessions with session timeline tracking for anti-hallucination. Stores memories, tracks session events, manages checkpoints, and provides grounding context.",
+    "version": "2.0.0",
+    "url": "http://localhost:8102",
+    "documentationUrl": "https://github.com/anthropics/claude-code",
+    "capabilities": {
+        "streaming": False,
+        "pushNotifications": False
+    },
+    "authentication": {
+        "schemes": []
+    },
+    "defaultInputModes": ["text"],
+    "defaultOutputModes": ["text"],
+    "skills": [
+        # ============================================================
+        # MEMORY SKILLS (Original)
+        # ============================================================
+        {
+            "id": "store_memory",
+            "name": "Store Memory",
+            "description": "Store a piece of information with semantic embedding for later retrieval. Supports different memory types: session, decision, code, chunk.",
+            "tags": ["memory", "storage", "embedding"],
+            "examples": [
+                "Store this code pattern for future reference",
+                "Remember this architectural decision"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "retrieve_memory",
+            "name": "Retrieve Memory",
+            "description": "Retrieve specific memories by ID or filter by type and session.",
+            "tags": ["memory", "retrieval"],
+            "examples": [
+                "Get memory with ID 42",
+                "List all decisions from this session"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "semantic_search",
+            "name": "Semantic Search",
+            "description": "Search through stored memories using natural language. Returns semantically similar content ranked by relevance.",
+            "tags": ["memory", "search", "semantic"],
+            "examples": [
+                "Find memories about authentication",
+                "Search for code patterns related to caching"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "summarize_session",
+            "name": "Summarize Session",
+            "description": "Store a comprehensive session summary including key decisions and code patterns discovered during the session.",
+            "tags": ["memory", "session", "summary"],
+            "examples": [
+                "Summarize this coding session",
+                "Store session digest with key decisions"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # TIMELINE SKILLS (New - Anti-Hallucination)
+        # ============================================================
+        {
+            "id": "timeline_log",
+            "name": "Timeline Log",
+            "description": "Log an event to the session timeline. Events include: user_request, decision, action, observation, error, checkpoint. Tracks causal chains and entities.",
+            "tags": ["timeline", "logging", "events", "grounding"],
+            "examples": [
+                "Log a decision to use React over Vue",
+                "Record an action taken on a file"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "timeline_get",
+            "name": "Timeline Get",
+            "description": "Retrieve timeline events for a session. Can filter by event type, get anchors only, or events since a specific ID.",
+            "tags": ["timeline", "retrieval", "events"],
+            "examples": [
+                "Get the last 20 events in this session",
+                "Get all decisions made so far"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "timeline_search",
+            "name": "Timeline Search",
+            "description": "Semantic search across timeline events. Find events related to a specific topic or query.",
+            "tags": ["timeline", "search", "semantic"],
+            "examples": [
+                "Search timeline for authentication changes",
+                "Find events related to database work"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "timeline_auto_detect",
+            "name": "Timeline Auto-Detect",
+            "description": "Auto-detect and log decisions/observations from a response text. Uses pattern matching to find implicit decisions and observations.",
+            "tags": ["timeline", "auto", "detection"],
+            "examples": [
+                "Analyze Claude's response for decisions",
+                "Auto-log detected observations"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # STATE SKILLS (New)
+        # ============================================================
+        {
+            "id": "state_get",
+            "name": "State Get",
+            "description": "Get current session state including goal, entity registry, pending questions, and checkpoint info.",
+            "tags": ["state", "session", "context"],
+            "examples": [
+                "Get current session state",
+                "What's the current goal?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "state_update",
+            "name": "State Update",
+            "description": "Update session state. Set current goal, add entities to registry, manage pending questions, add decisions.",
+            "tags": ["state", "session", "update"],
+            "examples": [
+                "Set current goal to fix auth bug",
+                "Register auth.py as the auth_file entity"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "state_init_session",
+            "name": "Initialize Session",
+            "description": "Initialize or resume a session for a project. Handles 4-hour session boundaries automatically.",
+            "tags": ["state", "session", "init"],
+            "examples": [
+                "Start or resume session for this project",
+                "Initialize session context"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # CHECKPOINT SKILLS (New)
+        # ============================================================
+        {
+            "id": "checkpoint_create",
+            "name": "Checkpoint Create",
+            "description": "Create a checkpoint snapshot of the current session state. Captures key facts, decisions, entities, and pending items.",
+            "tags": ["checkpoint", "snapshot", "save"],
+            "examples": [
+                "Create a checkpoint before major changes",
+                "Save session progress"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "checkpoint_load",
+            "name": "Checkpoint Load",
+            "description": "Load context from a checkpoint for session resumption. Returns grounding summary with goal, facts, decisions.",
+            "tags": ["checkpoint", "load", "resume"],
+            "examples": [
+                "Load the last checkpoint",
+                "Resume from previous session"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "checkpoint_list",
+            "name": "Checkpoint List",
+            "description": "List all checkpoints for a session. Shows checkpoint summaries and timestamps.",
+            "tags": ["checkpoint", "list"],
+            "examples": [
+                "List all checkpoints in this session",
+                "Show checkpoint history"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # GROUNDING SKILLS (New - Anti-Hallucination)
+        # ============================================================
+        {
+            "id": "context_refresh",
+            "name": "Context Refresh",
+            "description": "Pre-response grounding check. Returns current goal, entity registry, recent events, anchors, decisions, and potential contradictions. CALL THIS BEFORE COMPLEX RESPONSES.",
+            "tags": ["grounding", "context", "anti-hallucination"],
+            "examples": [
+                "Ground myself before responding",
+                "Check context before file changes"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "check_contradictions",
+            "name": "Check Contradictions",
+            "description": "Check if a statement contradicts known facts or decisions. Returns potential conflicts with established anchors and decisions.",
+            "tags": ["grounding", "verification", "anti-hallucination"],
+            "examples": [
+                "Check if this contradicts earlier decisions",
+                "Verify statement against known facts"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "verify_entity",
+            "name": "Verify Entity",
+            "description": "Verify an entity reference against the registry. Use when referencing files, variables, or other entities to ensure correctness.",
+            "tags": ["grounding", "entity", "verification"],
+            "examples": [
+                "Verify the auth_file entity",
+                "Check if this is the right file"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "mark_anchor",
+            "name": "Mark Anchor",
+            "description": "Mark a statement as a verified anchor fact. Anchors are high-confidence facts that should not be contradicted.",
+            "tags": ["grounding", "anchor", "fact"],
+            "examples": [
+                "Establish this as a verified fact",
+                "Mark this decision as final"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # CLAUDE.MD MANAGEMENT SKILLS
+        # ============================================================
+        {
+            "id": "claude_md_read",
+            "name": "Read CLAUDE.md",
+            "description": "Read the CLAUDE.md instructions file. Can read entire file or a specific section.",
+            "tags": ["claude_md", "instructions", "read"],
+            "examples": [
+                "Read my CLAUDE.md file",
+                "Show the Memory System section"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "claude_md_add_section",
+            "name": "Add CLAUDE.md Section",
+            "description": "Add a new section to CLAUDE.md. Use this to create new instruction categories.",
+            "tags": ["claude_md", "instructions", "write"],
+            "examples": [
+                "Add a new section for testing guidelines",
+                "Create an API conventions section"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "claude_md_update_section",
+            "name": "Update CLAUDE.md Section",
+            "description": "Update an existing section in CLAUDE.md. Can replace, append, or prepend content.",
+            "tags": ["claude_md", "instructions", "write"],
+            "examples": [
+                "Update the Memory System section",
+                "Append to the debugging guidelines"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "claude_md_add_instruction",
+            "name": "Add CLAUDE.md Instruction",
+            "description": "Add a single instruction/rule to a section. Creates the section if it doesn't exist.",
+            "tags": ["claude_md", "instructions", "write"],
+            "examples": [
+                "Add a rule about file naming",
+                "Add an instruction to the debugging section"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "claude_md_list_sections",
+            "name": "List CLAUDE.md Sections",
+            "description": "List all sections in the CLAUDE.md file.",
+            "tags": ["claude_md", "instructions", "list"],
+            "examples": [
+                "What sections are in my CLAUDE.md?",
+                "List all instruction categories"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "claude_md_suggest",
+            "name": "Suggest CLAUDE.md Updates",
+            "description": "Suggest CLAUDE.md additions based on session learnings. Analyzes anchors and decisions to recommend persistent instructions.",
+            "tags": ["claude_md", "instructions", "suggest"],
+            "examples": [
+                "What should I add to CLAUDE.md from this session?",
+                "Suggest instructions based on our work"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # VERIFICATION SKILLS (Anti-Hallucination)
+        # ============================================================
+        {
+            "id": "best_of_n_verify",
+            "name": "Best-of-N Verification",
+            "description": "Run a query N times and check consistency. Inconsistent outputs indicate potential hallucination. Use for critical facts.",
+            "tags": ["verification", "anti-hallucination", "consistency"],
+            "examples": [
+                "Verify this fact with multiple runs",
+                "Check if this answer is consistent"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "extract_quotes",
+            "name": "Extract Quotes",
+            "description": "Extract exact, word-for-word quotes from a document relevant to a query. Forces verbatim grounding instead of paraphrasing.",
+            "tags": ["verification", "grounding", "quotes"],
+            "examples": [
+                "Extract relevant quotes from this document",
+                "Find exact quotes about authentication"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "require_grounding",
+            "name": "Require Grounding",
+            "description": "Verify that a statement is grounded in stored facts (anchors or memories) before accepting it. Returns grounding sources or warning.",
+            "tags": ["verification", "grounding", "anti-hallucination"],
+            "examples": [
+                "Is this statement grounded in our facts?",
+                "Verify this claim against stored knowledge"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
+        # MOLTBOT-INSPIRED SKILLS (Human-Readable Transparency)
+        # ============================================================
+        {
+            "id": "daily_log_append",
+            "name": "Append to Daily Log",
+            "description": "Append an entry to today's daily log. Creates YYYY-MM-DD.md file in .claude/memory/. Entry types: decision, accomplishment, note, error.",
+            "tags": ["daily_log", "transparency", "moltbot"],
+            "examples": [
+                "Log a decision to today's log",
+                "Record an accomplishment"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "daily_log_append_session",
+            "name": "Append Session Summary to Daily Log",
+            "description": "Append a full session summary to the daily log. Includes decisions, accomplishments, errors solved, and notes.",
+            "tags": ["daily_log", "session", "transparency"],
+            "examples": [
+                "Add session summary to daily log",
+                "Log session end details"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "daily_log_read",
+            "name": "Read Daily Logs",
+            "description": "Read recent daily logs. Returns combined content from today and previous days.",
+            "tags": ["daily_log", "read", "transparency"],
+            "examples": [
+                "Show today's activity log",
+                "Read last 2 days of logs"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "daily_log_highlights",
+            "name": "Get Daily Log Highlights",
+            "description": "Get the most important entries from today's log for context injection. Returns decisions and accomplishments.",
+            "tags": ["daily_log", "highlights", "context"],
+            "examples": [
+                "Get today's highlights",
+                "What did we accomplish today?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "sync_memory_md",
+            "name": "Sync MEMORY.md",
+            "description": "Sync high-importance memories to MEMORY.md. Includes anchors, key decisions (importance >= 7), proven patterns (success >= 3).",
+            "tags": ["memory_md", "sync", "transparency"],
+            "examples": [
+                "Update MEMORY.md with important facts",
+                "Sync core knowledge to markdown"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "read_memory_md",
+            "name": "Read MEMORY.md",
+            "description": "Read the MEMORY.md core facts file. Contains anchors, key decisions, proven patterns, and preferences.",
+            "tags": ["memory_md", "read", "transparency"],
+            "examples": [
+                "Show core facts from MEMORY.md",
+                "What are the established facts?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "get_memory_md_summary",
+            "name": "Get MEMORY.md Summary",
+            "description": "Get a condensed summary of MEMORY.md for context injection. Returns essential facts in compact format.",
+            "tags": ["memory_md", "summary", "context"],
+            "examples": [
+                "Get core facts summary",
+                "Load MEMORY.md highlights"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "add_memory_md_fact",
+            "name": "Add Fact to MEMORY.md",
+            "description": "Add a fact directly to MEMORY.md. Sections: anchors, decisions, preferences.",
+            "tags": ["memory_md", "write", "transparency"],
+            "examples": [
+                "Add a verified fact to MEMORY.md",
+                "Record a key decision"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "check_flush_needed",
+            "name": "Check Flush Needed",
+            "description": "Check if pre-compaction flush is needed. Triggers if events > 50 or time > 30min since last flush.",
+            "tags": ["flush", "compaction", "transparency"],
+            "examples": [
+                "Check if we need to flush",
+                "Is memory export needed?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "pre_compaction_flush",
+            "name": "Pre-Compaction Flush",
+            "description": "Export all important session data to a markdown file before context loss. Creates flush_YYYYMMDD_HHMMSS.md.",
+            "tags": ["flush", "compaction", "transparency"],
+            "examples": [
+                "Export session data before compaction",
+                "Create memory flush"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "list_flushes",
+            "name": "List Memory Flushes",
+            "description": "List available memory flush files for a project.",
+            "tags": ["flush", "list", "transparency"],
+            "examples": [
+                "Show memory flush history",
+                "List all flushes"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "read_flush",
+            "name": "Read Memory Flush",
+            "description": "Read a specific memory flush file or the most recent one.",
+            "tags": ["flush", "read", "transparency"],
+            "examples": [
+                "Read the last flush",
+                "Show flush from earlier today"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        }
+    ]
+}