self-improvement

Files (1)

• SKILL.md

SKILL.md (first lines)

# Self Improvement Loop

Run after trading sessions and at end of day:

```bash
python3 ${OPENCLAW_HOME:-$HOME/.openclaw}/tools/risk/self_improve.py
```

Reads journal tags:
- `SETUP:`
- `RESULT:` WIN/LOSS/BREAKEVEN
- `R_MULTIPLE:`
- `ERROR:`

Writes:
- `trading/learning/latest-summary.json`
- `trading/learning/latest-summary.md`

Use output to update sizing, setup selection, and process rules.

Weekly cadence: follow the *Keep / Stop / Test* loop in
`knowledge/wiki-highlights/JOURNAL.md`. Reference
`knowledge/wiki-highlights/SIZING.md` when reviewing R-multiple
distribution and expectancy by setup, and
`knowledge/wiki-highlights/PSYCHOLOGY.md` when a rule violation is
the root cause rather than the strategy.

Bias-pattern aggregation (weekly): the `eod-review` skill tags every
losing trade and rule violation with the Munger tendencies that fired
(see `knowledge/wiki-highlights/MUNGER.md`). The weekly job here is
to look at the *aggregate* — not which biases fired on which trade,
but which biases keep firing across trades. Specifically:

- **A single tendency firing repeatedly** = a personal weakness to
  build a process antidote against. Example: if #14 Deprival
  Superreaction shows up on 4 of 7 losers this week, the antidote is
  not "try harder" but a *mechanical rule* that fires before you
  feel anything (e.g., "any position down 1.5R is closed without
  human intervention").
- **Lollapalooza events** (#25 — multiple tendencies firing on the
  same trade) decay slowest and damage most. Track them as their
  own column. Two lollapaloozas in a week is a *standdown* signal,
  even if total PnL is fine.
- **Errors of omission** — Munger and Buffett both stress that
  errors of omission (good trades you didn't take) typically dwarf
  errors of commission. The weekly review should explicitly list 1-3
  setups you saw but did not act on, what tendency stopped you, and
  what would have happened. Don't double-count regret, but don't
  pretend the misses didn't happen either.

The output of this aggregation feeds the *Stop* and *Test* lists in
the weekly Keep/Stop/Test loop. The *Keep* list comes from the
process scorecard; the *Stop* and *Test* lists come from the bias
aggregation.

From matiasvillaverde/openclaw-trader, MIT.

More from matiasvillaverde/openclaw-trader