profitability-signal

name	Profitability Signal
description	Execute the agent's Dune query, parse realized fee profitability into a structured signal in memory/profitability-signal.jsonl, and feed it to goal-review to adjust strategy
var
tags	["defi","analytics","self-improvement"]

Profitability Signal (weekly)

This is the measurement half of the profitability loop: it runs the agent's own Dune query, turns the rows into a structured signal sliced by time, and writes it to memory/profitability-signal.jsonl. goal-review reads that signal and turns it into one strategy recommendation. Over time the agent refines the query (scripts/dune/profitability.sql) and its strategy against measured outcomes — that closes the loop.

Honest by construction. This reports REALIZED fee income — which is ≈ $0 for a young agent. Never describe it as APY or projected yield. "No fees yet" is a valid, useful signal (recommendation: insufficient_data).

Setup (once per agent)

This skill needs two repo secrets:

• DUNE_API_KEY — the agent's Dune API key (dune.com → Settings → API).
• DUNE_PROFITABILITY_QUERY_ID — a saved Dune query id. Create an empty query in Dune, put its id here, then push the repo's query into it (Step 1). The query text lives in scripts/dune/profitability.sql and is the agent's to evolve.

Without these the skill is a safe no-op (it logs what it would do and exits 0).

Step 1 — (first run / after editing the query) push the query to Dune

node --env-file=.env --import tsx scripts/analyze-profitability.ts --write-query

This substitutes {{agent_wallet}} / {{window_days}} into scripts/dune/profitability.sql and writes it to the saved Dune query. Re-run this only when you've changed the SQL — not every cycle.

Step 2 — dry-run

node --env-file=.env --import tsx scripts/analyze-profitability.ts

Confirms which query it will execute and where it will write. No Dune credits spent, nothing written.

Step 3 — execute + write the signal

node --env-file=.env --import tsx scripts/analyze-profitability.ts --live --period=weekly

This executes the saved query (free-tier; ~2–3 Dune credits), parses the rows, and appends one signal object to memory/profitability-signal.jsonl:

{ "date": "...", "period": "weekly", "feesUsd": 0, "ilUsd": 0, "netPnlUsd": 0,
  "feeAprPct": null, "inRangePct": null, "recommendation": "insufficient_data",
  "rationale": "no realized fee income yet", "source": { "duneQueryId": 123, "rowCount": 0, "shape": "timeslice" } }

The script self-limits: it skips a re-run inside the period window (cost cap). Pass --force only to override deliberately.

Step 4 — log and act on the recommendation

Write a one-liner to memory/logs/${today}.md:

profitability-signal: fees $X | net $Y | rec=<recommendation> (<rationale>)

Then act on the recommendation (this is where strategy meets measurement):

• reposition / reduce_exposure → flag for the next lp-monitor run, or run it now.
• increase_liquidity → consider adding to the position on the next claim-diem.
• insufficient_data → no fees yet; keep accumulating, do not change strategy.
• hold → strategy is working; leave it.

goal-review will pick the signal up automatically on its weekly pass. If the same recommendation persists for 3+ cycles, consider whether the query itself needs to measure something better — edit scripts/dune/profitability.sql and re-run Step 1.

Step 5 — notify (optional)

./notify "profitability: net $Y over <window>d | rec=<recommendation>"

Cost note

Free-tier Dune execution only; one execution per period. The query window (--window, default 30 days) bounds the result size. Keep enabled: false until the agent has LP positions worth measuring.