memory_write(target="commitments/positions.md", content="# Current Positions\n\nMaintain your positions here. The agent reads this to score signal relevance.\n\n## Format\n\n- TICKER: SIZE, entry PRICE, thesis: BRIEF_THESIS\n\nExample:\n- AAPL: 500 shares, entry $175, thesis: AI integration undervalued\n- SPY Apr 520P: 10 contracts, thesis: hedging macro risk\n\n## Positions\n\n(Add your positions here)", append=false)

memory_write(target="commitments/trade-journal/README.md", content="Daily trade journal entries. Each file: commitments/decisions/<date>-<slug>.md with outcome tracking.", append=false)

mission_create(
  name: "commitment-triage",
  goal: "Trader triage. Read commitments/README.md for schema. Read commitments/positions.md for current positions. Priority order: (1) Position-relevant signals — any signal mentioning a ticker in the positions list gets immediacy=realtime and urgency=critical. Broadcast immediately via message tool. (2) Contradictory signal detection — if two pending signals point in opposite directions on the same ticker or thesis, flag as CONFLICT and surface both together immediately. (3) Market signals expire after 4 hours during market days, 24 hours otherwise. Research/thesis signals expire after 48 hours. (4) Route market intelligence (analyst reports, macro data) to intelligence destination via MemoryDoc. (5) Check decisions older than 7 days without outcome — prompt for outcome assessment. (6) Append triage summary to commitments/triage-log.md. (7) Alert on any position-relevant or conflicting signals.",
  cadence: "0 8,10,12,14,16,18 * * 1-5"
)

mission_create(
  name: "commitment-digest",
  goal: "Trader digest. Read commitments/README.md for schema. Read commitments/positions.md for current positions. If this is a morning run: (1) POSITION STATUS — list each position with any relevant signals from the last 24h. (2) OPEN RESEARCH — commitments tagged 'research' or 'thesis'. (3) PENDING DECISIONS — items with resolution_path=needs_decision. (4) CONFLICTING SIGNALS — any unresolved conflicts. If this is an evening run: (1) Summarize today's decisions from commitments/decisions/ with today's date. (2) For each decision, note if outcome data is available. (3) Prompt: 'Any trades to journal? Any thesis updates?' End with 'Did I miss anything?' Send via message tool.",
  cadence: "0 7,18 * * 1-5"
)

mission_create(
  name: "trader-weekly-review",
  goal: "Weekly trading review. Read all files in commitments/decisions/ from the past 7 days. For each decision: (1) What was decided and why. (2) If outcome data exists, was it positive or negative? (3) Which signals informed the decision — were those signal sources reliable? Also read commitments/positions.md — for each position, has the original thesis changed based on this week's signals? Flag any position where contradictory evidence has accumulated. Write review summary to context/intel/weekly-review-<date>.md as durable intelligence. Send via message tool.",
  cadence: "0 10 * * 6"
)

memory_write(
  target: "commitments/calibration.md",
  content: "# Trader Calibration\n\n- Always read commitments/positions.md before scoring signal relevance — a headline about AAPL is noise unless you hold AAPL\n- Position-relevant signals get immediacy=realtime — broadcast immediately, do not wait for digest\n- Market signals expire after 4 hours on trading days; research signals after 48 hours\n- When two signals contradict on the same ticker or thesis, flag as CONFLICT — never surface them independently\n- Trade decisions go in commitments/decisions/ with the standard schema, plus outcome tracking\n- Prompt for outcome assessment on decisions older than 7 days: 'You decided X a week ago. How did it play out?'\n- Pre-market brief leads with position-relevant signals; post-market prompt leads with today's decisions\n- Weekly review on Saturday assesses signal source reliability and thesis drift — write to context/intel/ as durable intelligence\n- The user maintains positions.md manually — do not modify it, only read it\n- Start conservative: surface everything, ask before acting on agent_can_handle items",
  append: false
)