journal-entry-writer

name	journal-entry-writer
description	Schema-aware workflow for writing structured journal entries. Ensures all entries conform to the Pydantic models in memory/schema.py. Use when the Executor needs to record an action, outcome, learning, or any other journal section. Activates after task execution, belief changes, or reflective processes.
compatibility	google-adk
metadata	{"scope":"executor","triggers":["after executing an approved action","after receiving Judge verdict","after dream cycle completes","during reflective mode","when learnings or connections are identified"],"requires_tools":["append_to_state","write_journal_entry"]}

Journal Entry Writer

You are activating the Journal Entry Writer skill. This ensures all journal entries are structured, schema-compliant, and complete.

Required Workflow

Steps 1-2 are ALWAYS mandatory. Steps 3-7 are conditional.

Step 1: Record in State (ALWAYS)

Use append_to_state to record the action/event in session state before writing to the journal. This ensures the state is available for other agents in the same session.

Key mappings:

• Action executed → key execution_log
• Judge verdict received → key judge_verdict
• Dream cycle completed → key dream_output
• Learning identified → key learnings

Step 2: Write Journal Entry (ALWAYS)

Use write_journal_entry tool with the appropriate section and structured content.

Available sections: dreams, judgments, executions, learnings, belief_mutations, open_questions, connections

Step 3: Record Learnings (if applicable)

If the execution produced a learning — something that changes how Jarvis should approach similar situations in the future — record it.

{
  "content": "<specific, actionable learning>",
  "source_mode": "awake|sleep|reflective"
}

Quality rule: A learning must be specific enough to change future behavior.

• Bad: "Error handling is important"
• Good: "The Stripe API returns 429 with retry-after header — always check it before retrying"

Step 4: Record Belief Mutations (if applicable)

If the Judge issued a belief mutation during this cycle, record it in the journal.

{
  "timestamp": "<ISO 8601>",
  "mutation_type": "new|updated|superseded|merged|coexisting",
  "belief": "<the belief statement>",
  "strength": 0.5,
  "reason": "<evidence-linked reason>"
}

Step 5: Record Open Questions (if applicable)

If the interaction raised questions that weren't answered, record them for future investigation.

Format: Plain text strings. Each question should be specific and investigable.

• Bad: "How does memory work?"
• Good: "Does Mem0's hybrid retrieval handle multi-language content for J.D.'s Spanish/English mix?"

Step 6: Record Connections (if applicable)

If this interaction connects to previous entries, past decisions, or dream seeds, record the connection.

Format: Plain text strings describing the connection.

• "This error pattern is the same one from 2026-03-10 — the retry logic fix applies here too"
• "The Dreamer's [INVERTED] seed about self-indexing memory is exactly what we implemented today"

Step 7: Validate Completeness

Before finishing, verify:

• Section name is valid (one of the 7 allowed sections)
• Content matches the JSON schema for structured sections
• Timestamps are ISO 8601 format
• No empty required fields
• Learnings are specific and actionable (if present)
• Belief mutations have all required fields (if present)

JSON Schemas by Section

dreams (DreamCycle)

{
  "timestamp": "2026-03-13T10:00:00",
  "seed": "the initial dream seed",
  "branches": ["branch 1", "branch 2"],
  "depth_reached": 3,
  "breakthrough": "the breakthrough insight or null"
}

judgments (Judgment)

{
  "timestamp": "2026-03-13T10:00:00",
  "action": "what was being judged",
  "action_weight": 5,
  "verdict": "approved|modified|rejected",
  "reasoning": "why this verdict",
  "original": "original action if modified",
  "modified": "modified action if applicable"
}

executions (Execution)

{
  "timestamp": "2026-03-13T10:00:00",
  "action": "what was executed",
  "status": "completed|failed|partial",
  "outcome": "what happened",
  "artifacts": ["file.py", "output.log"]
}

learnings (Learning)

{
  "content": "specific actionable learning",
  "source_mode": "awake|sleep|reflective"
}

belief_mutations (BeliefMutation)

{
  "timestamp": "2026-03-13T10:00:00",
  "mutation_type": "new|updated|superseded|merged|coexisting",
  "belief": "the belief statement",
  "strength": 0.5,
  "reason": "evidence-linked reason"
}

open_questions

Plain text strings: "The specific question to investigate"

connections

Plain text strings: "Description of the connection to past experience"

See Also

• references/schema-reference.md — Pydantic models in human-readable format
• references/entry-examples.json — valid and invalid examples with annotations