Index of every page on this site: llms.txt
All skills
tradermonty avatar

/trader-memory-core

@f95e770
Run it yourself

Track investment theses across their lifecycle — from screening idea to closed position with postmortem. Register theses from screener outputs, manage state transitions, attach position sizing, review due dates, and generate postmortem reports with P&L and MAE/MFE analysis. Trigger when user says "register thesis", "track this idea", "thesis status", "review due", "close position", "postmortem", or "trading journal".

Use this Skill: https://skilld.dev/gh/tradermonty/claude-trading-skills/trader-memory-core

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈111 tokens for metadata: the name and description. ≈5.6k when used: this file. ≈6.9k more on demand in 4 files.

Description uses 5.5% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈441 characters, or 5.5%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Trader Memory Core

Overview

Persistent state layer that bundles screening → analysis → position sizing → portfolio management outputs into a single "thesis object" per investment idea. Tracks what you thought, what happened, and what you learned — across conversations.

Phase 1 supports single-ticker theses: dividend_income, growth_momentum, mean_reversion, earnings_drift, pivot_breakout.

When to Use

  • After a screener (kanchi, earnings-trade-analyzer, vcp, pead, canslim, edge-candidate-agent) produces candidates
  • When transitioning a thesis from IDEA → ENTRY_READY → ACTIVE → CLOSED
  • When attaching position-sizer output to a thesis
  • When checking which theses are due for review
  • When closing a position and generating a postmortem with lessons learned

Prerequisites

  • Python 3.10+
  • pyyaml (already in project dependencies)
  • jsonschema (already in pyproject.toml; required by thesis_store.py and every command that imports it, including thesis_ingest.py and thesis_review.py)
  • FMP API key (optional, only for MAE/MFE calculation in postmortem)

How to invoke the CLI

Use the stdlib-only launcher trader_memory_cli.py for all CLI work. It transparently routes through uv run --project <repo> when uv is available, so the repo's pinned jsonschema is reachable even from a foreign cwd or from python3 with no global jsonschema (e.g. cron / Hermes profile runs):

# From inside the repo
python3 skills/trader-memory-core/scripts/trader_memory_cli.py store --state-dir state/theses list

# From any other cwd (cron, profile, distribution runner) — point the launcher at the repo
export CLAUDE_TRADING_SKILLS_REPO=/path/to/claude-trading-skills
python3 "$CLAUDE_TRADING_SKILLS_REPO/skills/trader-memory-core/scripts/trader_memory_cli.py" \
  store --state-dir /path/to/state/theses list

Subcommands: store → thesis_store.py, ingest → thesis_ingest.py, review → thesis_review.py. Everything after the subcommand is forwarded verbatim, so existing argument flags (--state-dir, transition, open-position, etc.) work unchanged.

If the launcher reports that jsonschema is not importable AND uv is not on PATH, the actionable fixes (in priority order) are:

  1. Install uv (https://docs.astral.sh/uv/) and re-run the launcher.
  2. Install the project's dependencies into the current interpreter:
    uv pip install -e /path/to/claude-trading-skills
    # or, as a last resort:
    python3 -m pip install jsonschema

Do not treat the thesis store as unavailable and do not mutate state/theses/*.yaml by hand to work around a missing dependency — schema validation is part of thesis state integrity.

Equity accounting is long-only. Do not ingest parabolic-short plans or record short-equity trades through manual; there is no adapter for those plans. Explicit SHORT directions are rejected before they can be discarded by raw registration, manual ingest, equity opening or shares-report attachment. Generic registration does not import a raw position: use validated futures attachment or open-position --contracts for futures LONG/SHORT.

When both initial prices are supplied, they must be finite numeric/non-bool values. Positive initial stop > positive entry is rejected at manual ingest, shares-report attachment and default equity activation (using the actual fill). Absent optional pairs and existing finite nonpositive-price behavior remain compatible. Do not apply this initial-entry rule to an ACTIVE/PARTIALLY_CLOSED long's stop: a profitable long may update its stop above original entry. Manual source prices remain provenance only; authoritative fills still come from open-position. For a backfill with a profit-protecting stop, open the long with its initial stop, then update the current stop after activation.

For a real long that gaps and fills below its planned stop, keep the default refusal unless explicitly acknowledging the actual fill with open-position --allow-gap-fill (API: allow_gap_fill=True). This opt-in only bypasses the positive fill-below-positive-stop geometry guard for equity; SHORT intent, invalid prices/quantities/state and all ingest/attachment validation still refuse. Futures reject this equity-only option. A normal/equal fill adds no gap annotation. The flag must be a Python bool in API calls. The acknowledgement still requires a share count: pass --shares or attach a sized position.

Preserve the target, planned stop, sizing audit fields and provenance. Store the actual fill and append a gap fill acknowledged entry to the protected ACTIVE status_history.reason, including exact actual fill and stop values. A supplied reason must remain a string. This audit survives later stop changes; inspect status history when reviewing the gap. Closing/trimming uses the actual fill for long-side P&L. This records a fill that already occurred, not approval to open a new trade or retain a breached stop. Planned sizing fields are not recalculated by this acknowledgement.

python3 skills/trader-memory-core/scripts/trader_memory_cli.py store \
  --state-dir state/theses/ open-position <id> \
  --actual-price 88 --actual-date 2026-11-01 --shares 10 \
  --allow-gap-fill --reason "record actual broker fill"

Explicit legacy short equity records (a SHORT word or a bare SELL, such as SELL_SHORT, Short position or SELL; labels such as Sell to close stay long) are refused before new accounting or MAE/MFE calculations. CLOSED/INVALIDATED legacy shorts can produce historical-only postmortems: P&L, MAE/MFE, position valuation and risk are shown as unavailable (—), with an explicit unsupported-short explanation. This report fetches no prices and leaves thesis/index state, stored outcomes and provenance unchanged; old values are not validated or corrected. ACTIVE/PARTIALLY_CLOSED records still cannot generate postmortems. They can still be marked reviewed or terminated as INVALIDATED without an exit price; these bookkeeping actions preserve all existing monetary fields and provenance. Summary statistics and monthly reports skip explicit shorts and report their excluded count (excluded_unsupported_equity_shorts in summary JSON when nonzero). Old manual provenance was free text: BUY, Long position and other non-short labels retain legacy long accounting. New direction inputs remain strictly LONG. Read-only inspection remains available; no old outcomes are migrated or re-signed. Ambiguous legacy records cannot be classified from a trailing stop alone. Keep short-equity journals external; never relabel one as a futures contract.

Workflow

1. Register — Ingest screener output as thesis

Read the screener's JSON output and convert to thesis using the appropriate adapter.

python3 skills/trader-memory-core/scripts/trader_memory_cli.py ingest \
  --source kanchi-dividend-sop \
  --input reports/kanchi_entry_signals_2026-03-14.json \
  --state-dir state/theses/

Supported sources: kanchi-dividend-sop, earnings-trade-analyzer, vcp-screener, pead-screener, canslim-screener, edge-candidate-agent, manual.

Each thesis starts in IDEA status.

For kanchi-dividend-sop, registration is fail-closed: each row must carry one of CLEAN-PASS, PASS-CAUTION, or CONDITIONAL-PASS in verdict. Missing verdicts and HOLD-REVIEW / STEP1-RECHECK / FAIL rows are skipped and never written to thesis state.

Manual brokerage entry (fractional shares)

For trades that did not come from a screener — e.g. fractional-share brokers (IBKR, Robinhood, IBI Smart, Alpaca, eToro) or hand journaling — use the manual source with a free-form JSON file (a single object or an array):

{
  "ticker": "AMD",
  "thesis_statement": "AMD AI accelerator momentum, fractional IBI Smart position",
  "thesis_type": "growth_momentum",
  "entry_price": 142.10,
  "entry_date": "2026-05-02",
  "shares": 7.86,
  "stop_price": 128.00
}
python3 skills/trader-memory-core/scripts/trader_memory_cli.py ingest \
  --source manual --input amd.json --state-dir state/theses/

Required: ticker, thesis_statement, thesis_type (one of dividend_income, growth_momentum, mean_reversion, earnings_drift, pivot_breakout). stop_price/stop_loss and target_price/take_profit map to exit.stop_loss/exit.take_profit; entry_price/entry_date/shares are kept in origin.raw_provenance — the authoritative entry price/date and share count are set when you open the position (below). shares may be fractional (the schema accepts any positive number). Like every adapter, manual ingest creates an IDEA thesis only — it never mutates status directly.

To record an already-open broker position, run the explicit lifecycle sequence (the --event-date flags backdate the history so it stays chronological):

# 1. ingest → IDEA (stamped at entry_date)
python3 .../trader_memory_cli.py ingest --source manual --input amd.json --state-dir state/theses/
# 2. IDEA → ENTRY_READY (backdated)
python3 .../trader_memory_cli.py store --state-dir state/theses/ transition <id> ENTRY_READY \
  --reason "existing IBI Smart position" --event-date 2026-05-02
# 3. ENTRY_READY → ACTIVE (fractional shares, backdated)
python3 .../trader_memory_cli.py store --state-dir state/theses/ open-position <id> \
  --actual-price 142.10 --actual-date 2026-05-02 --shares 7.86 --event-date 2026-05-02

2. Query — Search and list theses

python3 skills/trader-memory-core/scripts/trader_memory_cli.py store \
  --state-dir state/theses/ list --ticker AAPL --status ACTIVE

Filter by --ticker, --status, or --type.

3. Update — Transition, attach position, link reports

Each lifecycle operation is available both as a Python function and as a thesis_store.py CLI subcommand. --event-date / --actual-date accept a plain YYYY-MM-DD (widened to midnight UTC) or a full ISO timestamp.

State transition (IDEA → ENTRY_READY only):

python3 skills/trader-memory-core/scripts/trader_memory_cli.py store --state-dir state/theses/ \
  transition <id> ENTRY_READY --reason "validated" [--event-date YYYY-MM-DD]

--event-date backdates status_history.at (use it when backfilling an existing position so the later backdated open-position stays chronological). Python: thesis_store.transition(state_dir, thesis_id, "ENTRY_READY", reason, event_date=...).

Open position (ENTRY_READY → ACTIVE — the only path to ACTIVE):

python3 .../trader_memory_cli.py store --state-dir state/theses/ open-position <id> \
  --actual-price 142.10 --actual-date 2026-05-02 --shares 7.86 [--event-date 2026-05-02]

--shares is required for an equity thesis unless attach-position already recorded a share count; otherwise open-position exits 1 with ERROR: open_position() requires a share count ... and nothing is saved. ACTIVE theses recorded before this check without shares still close on the legacy per-share path. --shares accepts fractional quantities. Python: thesis_store.open_position(state_dir, thesis_id, actual_price, actual_date, shares=..., event_date=...). shares (and shares_remaining, when present) must be a finite, positive number no greater than 10<sup>12</sup> (a sanity bound, not an economic constraint — fractional shares below the cap remain unrestricted). NaN, ±Infinity, and absurdly large values (e.g. a malformed position-sizer report) are rejected with a clean error at save time, on open-position, attach-position, and trim alike.

For a futures thesis, use --contracts instead of --shares (see "Futures positions" below) — if attach-futures-position already populated the position, omit --contracts and only pass --actual-price/--actual-date.

store doctor validates file/index consistency, prints a JSON report, and exits 1 when ok is false or the state directory is missing or not a directory.

Trim — partial close (ACTIVE/PARTIALLY_CLOSED → PARTIALLY_CLOSED, or → CLOSED when the whole remainder is sold):

python3 .../trader_memory_cli.py store --state-dir state/theses/ trim <id> \
  --shares-sold 4 --price 120.00 --date 2026-05-10

position.shares is the original opened quantity (immutable); position.shares_remaining tracks what is still open. Each trim appends a status_history ledger entry (shares_sold / price / proceeds / realized_pnl). outcome.pnl_dollars is the cumulative realized P&L (Σ all trims + final close); outcome.pnl_pct = pnl_dollars / (entry_price × original_shares) × 100. A trim that sells the entire remainder closes the thesis (default exit_reason: manual, overridable with --exit-reason). --date is the ledger timestamp (override with --event-date). Python: thesis_store.trim(state_dir, thesis_id, shares_sold, price, date, ...).

Status invariants: ACTIVE ⇒ shares_remaining == shares; PARTIALLY_CLOSED ⇒ 0 < shares_remaining < shares; CLOSED ⇒ shares_remaining == 0. Legacy theses (no shares_remaining) are treated as fully open at runtime.

For a futures thesis, use --contracts-sold instead of --shares-sold — close/terminate need no flag changes; they read position.asset_type and dispatch automatically (see "Futures positions" below).

Close or invalidate (→ CLOSED or INVALIDATED):

python3 .../trader_memory_cli.py store --state-dir state/theses/ close <id> \
  --exit-reason target_hit --actual-price 165.00 --actual-date 2026-06-01
python3 .../trader_memory_cli.py store --state-dir state/theses/ terminate <id> \
  --terminal-status INVALIDATED --exit-reason "thesis broke"

close accepts an ACTIVE or PARTIALLY_CLOSED thesis; from PARTIALLY_CLOSED it adds the final leg and reports the cumulative outcome.

Python: thesis_store.terminate(state_dir, thesis_id, terminal_status, exit_reason, actual_price, actual_date). For CLOSED, delegates to close() which computes P&L (fractional-share aware). For INVALIDATED, P&L is computed if entry/exit prices are available.

Record review (any non-terminal):

Use thesis_store.mark_reviewed(state_dir, thesis_id, review_date=..., outcome="OK"|"WARN"|"REVIEW") to advance next_review_date and record alerts.

Attach position-sizer output:

python3 .../trader_memory_cli.py store --state-dir state/theses/ attach-position <id> \
  --report reports/position_report.json

Python: thesis_store.attach_position(state_dir, thesis_id, report_path) to link position sizing data. Validates that the report mode is "shares" (not budget).

Futures positions (contracts / multiplier / direction)

A thesis whose position.asset_type == "futures" (or quantity_unit == "contracts") is a futures thesis. Futures theses use quantity / quantity_remaining (whole contracts — no fractional contracts) instead of shares / shares_remaining, carry a direction (LONG or SHORT) and a multiplier, and every P&L computation (close, terminate, trim) applies (exit_price - entry_price) × multiplier × quantity × sign (sign = +1 LONG, −1 SHORT) instead of the equity per-unit formula. close / terminate / trim / open-position all dispatch on position.asset_type automatically — no separate futures subcommands for those four operations. USD-denominated contracts only — there is no FX conversion in the P&L path, so a non-USD contract_spec.currency is rejected outright rather than computing P&L in the wrong currency's magnitude.

Attach a futures-position-sizer SIZED report (step 6 of the Shapiro contrarian pipeline — futures-position-sizer → trader-memory-core):

python3 .../trader_memory_cli.py store --state-dir state/theses/ \
  attach-futures-position <id> --report reports/futures_position_es_2026-05-10.json

Rejects a NO_TRADE report (sizing_status != "SIZED"), an invalid direction, a non-positive/fractional contracts count, a non-finite/non-positive contract_spec.multiplier, or a non-USD contract_spec.currency. Re-attach status guard is IDEA/ENTRY_READY only — stricter than equity's attach-position (which also allows ACTIVE): re-attaching a futures position on ACTIVE would silently overwrite the entire position dict including direction, flipping the sign of every subsequent P&L computation. Correcting an already-open futures position needs a fresh thesis (or a future dedicated "amend" operation) — not a re-attach.

Direct open, no attach (build the position from CLI flags instead of a SIZED report — --contract-currency is required here since there is no contract_spec to read a currency from, and must be USD):

python3 .../trader_memory_cli.py store --state-dir state/theses/ open-position <id> \
  --actual-price 5000 --actual-date 2026-05-10 \
  --contracts 2 --multiplier 50 --direction SHORT --contract-symbol ES \
  --contract-currency USD

Trim / close / terminate — same subcommands as equity, --contracts-sold in place of --shares-sold:

python3 .../trader_memory_cli.py store --state-dir state/theses/ trim <id> \
  --contracts-sold 1 --price 4950.00 --date 2026-05-12
python3 .../trader_memory_cli.py store --state-dir state/theses/ close <id> \
  --exit-reason target_hit --actual-price 4900.00 --actual-date 2026-05-15

Python: thesis_store.attach_futures_position(state_dir, thesis_id, report_path), thesis_store.open_position(state_dir, thesis_id, actual_price, actual_date, contracts=..., multiplier=..., direction=...).

Link related reports:

Use thesis_store.link_report(state_dir, thesis_id, skill, file, date) to cross-reference analysis documents.

4. Review — Check due dates and monitoring status

python3 skills/trader-memory-core/scripts/trader_memory_cli.py review \
  --state-dir state/theses/ review-due --as-of 2026-04-15

List theses with next_review_date <= as_of. Use with kanchi-dividend-review-monitor triggers (T1-T5) for systematic review.

5. Postmortem — Close and reflect

python3 skills/trader-memory-core/scripts/trader_memory_cli.py review \
  --state-dir state/theses/ postmortem th_aapl_div_20260314_a3f1

Generate a structured postmortem in state/journal/. If FMP API key is available, includes MAE/MFE (Maximum Adverse/Favorable Excursion) metrics.

Record the lessons learned on a CLOSED or INVALIDATED thesis with record-lessons. This produces the lessons_log_entry of the trade-memory-loop lessons step; the text lands in outcome.lessons_learned.

python3 skills/trader-memory-core/scripts/trader_memory_cli.py store \
  --state-dir state/theses/ record-lessons th_aapl_div_20260314_a3f1 \
  --text "Sized too large into earnings; cut the position before the print."

Use --file PATH (UTF-8) instead of --text for multi-line text, and --text="..." when the text starts with -. If a lesson already exists the command refuses; pass --append to add a line or --replace to overwrite. Keep one plain paragraph per lesson, because lists and blank lines render as separate Markdown blocks in the monthly report. Re-run review postmortem afterwards to refresh the journal report; review monthly-report --month YYYY-MM lists the lessons under "Top Lessons".

Summary statistics:

python3 skills/trader-memory-core/scripts/trader_memory_cli.py review \
  --state-dir state/theses/ summary

Shows win rate, average P&L%, and per-type breakdown across all closed theses.

Output Format

Thesis YAML (state/theses/)

Each thesis is a YAML file with:

  • Identity: thesis_id, ticker, created_at
  • Classification: thesis_type, setup_type, catalyst
  • Lifecycle: status, status_history
  • Entry/Exit: target prices, actual prices, conditions
  • Position: shares (fractional supported), value, risk (from position-sizer or open-position --shares); or, for futures, quantity/multiplier/direction/contract_spec (from futures-position-sizer or open-position --contracts)
  • Monitoring: review dates, triggers, alerts
  • Origin: source skill, screening grade, raw provenance
  • Outcome: P&L, holding days, MAE/MFE, lessons learned

Index (state/theses/_index.json)

Lightweight index for fast queries without loading full YAML files.

Journal (state/journal/)

Postmortem markdown reports: pm_{thesis_id}.md.

Supported Python Read and Validation API

Use these public functions from scripts/thesis_store.py for replay and other Python consumers:

from pathlib import Path
import thesis_store

thesis = thesis_store.get(Path("state/theses"), thesis_id)
thesis_store.validate_thesis(thesis)
  • get(state_dir, thesis_id) loads fresh YAML data without modifying the thesis file or index. It does not validate the loaded content. Missing files raise FileNotFoundError; malformed YAML raises yaml.YAMLError.
  • validate_thesis(thesis) checks the JSON Schema and business invariants, returns None on success, and raises ValueError on validation failure. It does not mutate the input or write state.
  • Validate loaded or normalized records explicitly before relying on their schema and business invariants. Keep consumers on these supported functions; underscore-prefixed helpers are internal implementation details.

Key Principles

  • Forward-only transitions: IDEA → ENTRY_READY → ACTIVE → CLOSED (no backtracking)
  • Raw provenance: All original screener data preserved in origin.raw_provenance
  • Atomic writes: All file operations use tempfile + os.replace
  • Git-tracked state: state/ directory is committed, providing audit trail
  • Phase 1 scope: Single-ticker theses only (pair trades and options in Phase 2)

Resources

  • references/thesis_lifecycle.md — Status states and valid transitions
  • references/field_mapping.md — Source skill → canonical field mapping
  • schemas/thesis.schema.json — JSON Schema for thesis validation
  • ../../examples/workflows/trade-memory-loop/sample-run-full-path/ — Worked end-to-end Plan → Trade → Record → Postmortem → Backtest → Journal example

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning5d4 checks · Risk MEDIUM
  • Gen Agent Trust Hub5d

    The skill manages investment thesis data and utilizes dynamic code execution to assemble its internal logic from various file fragments. It also performs command execution through a launcher script to manage different CLI subcommands and fetches market data from a third-party financial API.

  • Socket5d

    No alerts

  • Snyk5d

    Risk: LOW · No issues

  • ZeroLeaks7mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at f95e770. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated last week

README badge

README badge for tradermonty/claude-trading-skills/trader-memory-core