This directory is a read-only IB (Interactive Brokers) diagnostics toolchain:
pull account data from IB, land it in a local data lake, and turn it into a
graded portfolio-diagnostics report. Nothing here ever places, modifies, or
cancels an order. Ingestion connects with readonly=True; analysis never
touches the network.
This SKILL.md is the entry point. It does not run anything itself — it tells you
(and OpenClaw) which sub-skill to run, in what order, and how the pieces fit.
0. First-run setup
Before running any sub-skill, make sure a config exists. This index owns
onboarding; the sub-skills stay gated until config.yaml is present.
Detect. If .ib-suite/config.yaml already exists, config is ready —
skip to §2. Otherwise continue.
bash
test -d {baseDir}/.venv || bash {baseDir}/scripts/setup_venv.sh
Ask the user one question: connect to live (real account, port 4001)
or paper (simulated, port 4002)? Default is live — every connection
in this toolchain is readonly=True, so live is read-only too.
Generate the config with the chosen mode (defaults to live):
bash
{baseDir}/.venv/bin/python {baseDir}/scripts/init_config.py \
--mode live --out .ib-suite/config.yaml
It refuses to overwrite an existing config unless you add --force.
Report back: the config path and the resulting mode/port. Remind the
user to start IB Gateway with Read-Only API enabled before /ib-sync,
and that ib-trade-history and ib-dividend-income share one Flex token but
keep separate per-window Query ID maps in ignored local config (flex.token
with flex.trade_history_query_ids / flex.dividend_query_ids); never echo
either credential.
Proceed to §2 and run /ib-sync → /ib-analyze.
Runtime config and data stay workspace-local under <workspace>/.ib-suite/
(gitignored); the skill dir ships only code and config.example.yaml.
skills/ib-suite/
SKILL.md # <- you are here (index / router)
scripts/setup_venv.sh # shared venv bootstrap (installs ib-common editable)
ib-common/ # shared library (installed editable into .venv)
ib-gateway/ # /ib-sync : IB/Flex -> local data lake
ib-account-overview/ # /ib-account-overview: IB account -> financial overview (no persistence)
ib-positions-overview/ # /ib-positions-overview: IB positions -> enriched, ranked overview (no persistence)
ib-daily-pnl/ # /ib-daily-pnl: IB live P&L -> today's realized/unrealized, ranked (no persistence)
ib-trade-history/ # /ib-trade-history: Flex executions -> stdout JSON (no persistence)
ib-dividend-income/ # /ib-dividend-income: Flex dividends -> stdout JSON (no account-data persistence)
ib-options-overview/ # /ib-options-overview: IB live options -> Greeks and risk overview (no persistence)
ib-portfolio-analyst/ # /ib-analyze: data lake -> report.md + charts
Scope. In: read-only sync, snapshots, Parquet history, and P0–P3 findings
across account health, concentration, P&L attribution, trade review, portfolio
risk, pre-trade simulation, and Flex-only dividend income. Out (hard boundary): order
placement/modification/cancellation, live WhatIf margin checks, real-time market
data, and any write path to IB. Do not add these under the banner of
"completeness".
setup_venv.sh -> ib-gateway /ib-sync -> ib-portfolio-analyst /ib-analyze
(install ib-common) (write data lake) (read lake -> report)
Setup once (or after dependency changes). Bootstraps the shared .venv
and installs ib-common (editable) plus runtime deps. Idempotent.
bash
bash {baseDir}/scripts/setup_venv.sh
# config.yaml is created by "0. First-run setup" (writes .ib-suite/config.yaml)
Runtime data lives outside the skill dir. Keep the real config.yaml and
the data lake under the workspace, e.g. <workspace>/.ib-suite/config.yaml
and <workspace>/.ib-suite/data/ (set storage.root: .ib-suite/data). The
skill directory ships only code and config.example.yaml; reinstalling the
skill must never overwrite user data. Entry scripts take explicit --config
/ --out and don't depend on the current working directory.
Ingest (ib-gateway, online). Start IB Gateway (paper 4002 / live 4001)
with API access, then run /ib-sync. Writes data/snapshots/<account>/<ts>.json
and appends data/timeseries/positions_history.parquet. See
ib-gateway/SKILL.md.
Analyze (ib-portfolio-analyst, offline). Run /ib-analyze against a
snapshot to produce report.md + .html/.png charts. See
ib-portfolio-analyst/SKILL.md.
Which skill do I run?
You want to…
Run
Refresh account/position data from IB
ib-gateway → /ib-sync
See account equity, margin, liquidity & P&L right now
ib-account-overview → /ib-account-overview
List every position, ranked, with the most concentrated name
ib-positions-overview → /ib-positions-overview
See how the account did today and which names drove it
ib-daily-pnl → /ib-daily-pnl
List historical fills, commission, realized P&L and win/loss statistics
ib-trade-history → /ib-trade-history
See paid/expected dividends, tax, attribution, annual income and yield
ib-dividend-income → /ib-dividend-income
Configure shared Flex credentials or dividend query fields
Inspect option holdings, IV, Greeks, expiry exposure, and concentration
ib-options-overview → /ib-options-overview
Produce a diagnostic report from existing data
ib-portfolio-analyst → /ib-analyze
Test any skill without IB
its tests/ fixtures (see §5)
Note on optional inputs./ib-sync v1 lands only the account snapshot and
positions. Daily bars, executions, and dividends are optional JSON arrays
(matching the DailyBar / Execution / Dividend schema, e.g. exported from a
Flex report) that you pass to /ib-analyze directly; the corresponding report
sections appear only when their data is supplied.
3. Sub-skill module spec
Every functional sub-skill directory (ib-gateway, ib-portfolio-analyst, and
any future one) MUST follow this contract:
SKILL.md with valid frontmatter. Required name + description, and
metadata.openclaw (requires.bins, requires.config, os) so OpenClaw can
discover and gate it. name MUST equal the directory name (lowercase, stable).
description states the read-only boundary. Start with "Read-only", say
what it does and when it triggers, and stay narrow enough to avoid mis-firing.
scripts/ holds the deterministic entry logic. Use argparse; validate
required args; exit non-zero with an actionable message on failure. Commands in
SKILL.md use the {baseDir} placeholder and .venv/bin/python — never
absolute or hardcoded user paths.
Network/IB access hides behind an injectable factory (client_factory,
http_get) so tests/ run fully offline against fixtures/.
Reuse ib-common, don't fork it. Types come from ib_common.schema
(pydantic v2); reuse Finding, grade(), thresholds, storage helpers, and
render() for charts. New diagnostic modules live in ib_analyst/, expose
analyze(...) -> list[Finding] with a module-level DIM constant and an
optional build_chart(...) -> plotly.Figure, and are wired into
analyze.py's run(). New thresholds MUST also be added to
ib-common/config.example.yaml under thresholds:.
from __future__ import annotations at the top of every module; type-annotate
functions; docstring public functions.
4. Invocation
As OpenClaw slash commands (primary). Once discovered, the sub-skills expose
/ib-sync, /ib-account-overview, /ib-positions-overview, /ib-daily-pnl,
/ib-trade-history, /ib-dividend-income, /ib-options-overview, and
/ib-analyze. Gating is driven by each skill's metadata.openclaw (python3
on PATH, a config.yaml, and a supported OS).
As direct scripts (for automation / other systems). Call the entry scripts
with the shared interpreter; they print structured, parseable results (dicts /
output paths) to stdout and return non-zero on failure:
As a library.import ib_common (installed editable) for config/schema/
storage/metrics/charts. ib-trade-history and ib-dividend-income share the
Flex token but keep separate per-window Query ID maps (flex.token with
flex.trade_history_query_ids / flex.dividend_query_ids) in ignored
local config. The dividend skill requires numeric windows and its standalone
field/window guide is
ib-dividend-income/flex-query-setup.md.
Never hardcode or echo tokens, Query IDs, account numbers, or user paths.