nova-economy-contributor

name	nova-economy-contributor
version	0.2.0
description	Autonomous skill for the IronClaw Swarm Economy. Reads the latest MESA simulation output from /data/simulation/outputs/, uploads the interaction graph to NOVA via nova-submit, then reports the key economic signals. Fully autonomous — no user confirmation required.
author	Julien Carbonnell
tags	["nova","swarm","economy","simulation","near"]
activation	{"keywords":["nova-economy-contributor","nova economy contributor","swarm contribution","contribute swarm","economy contribution"],"patterns":["nova.economy.contributor","swarm.*contribut","contribut.*swarm"],"tags":["nova","swarm"],"max_context_tokens":3000}

nova-economy-contributor v0.2.0

FIRST ACTION REQUIRED — DO THIS BEFORE ANYTHING ELSE:

1. Call memory_read with path swarm/config.md — this gives you NOVA_ACCOUNT_ID and NOVA_GROUP_ID
2. Call memory_read with path swarm/last_contribution.md — this gives you the previous round number (optional, use 1 if not found)
3. Call time — get the current UTC timestamp Only after these three calls, proceed to Step 3 (build payload). Do NOT use the shell tool at any point.

Autonomous skill for the IronClaw Swarm Economy experiment. When activated:

1. Reads credentials from agent memory
2. Reads the latest simulation output from /data/simulation/outputs/
3. Uploads the graph JSON to NOVA via the nova-submit tool
4. Writes a summary to memory and reports key economic signals to the user

No user input or confirmation is required at any step.

---

CRITICAL CONSTRAINTS

• This skill is fully autonomous. Do NOT ask the user for confirmation, parameters, or approval at any step.
• Do NOT use the shell tool at any point. It is not available for reading environment or credentials.
• Credentials are stored in agent memory. Always start by reading swarm/config.md with memory_read.
• NOVA_API_KEY is a secret available in the process environment. Pass it directly as the api_key parameter of nova-submit without trying to read or echo it.
• The ONLY tool for NOVA upload is nova-submit. Do not attempt HTTP calls.
• Never report success unless nova-submit returned a CID (not "PENDING" or empty).
• If any step fails, continue to the next step and report all errors at the end.

---

Flow

Step 1 — Read config (MANDATORY FIRST CALL)

Call memory_read with path swarm/config.md.

Parse these values:

• NOVA_ACCOUNT_ID — e.g. ironclaw-swarm-agent1.nova-sdk-6.testnet
• NOVA_GROUP_ID — e.g. ironclaw-swarm-economy
• NOVA_API_KEY — the API key (starts with nova_sk_)

If swarm/config.md is not found or any value is missing, stop immediately and report:

✗ nova-economy-contributor: swarm/config.md not found or incomplete.
  Required fields: NOVA_ACCOUNT_ID, NOVA_GROUP_ID, NOVA_API_KEY

Step 2 — Read last contribution (optional)

Call memory_read with path swarm/last_contribution.md.

Parse the round: field. If not found, use round = 0 (the simulation file will determine the actual round number).

Call the time tool to get the current UTC timestamp.

Step 3 — Read the latest simulation output

Derive the agent's short name from NOVA_ACCOUNT_ID by taking everything before the first dot. Example: ironclaw-swarm-agent1.nova-sdk-6.testnet → ironclaw-swarm-agent1.

Call the shell tool with:

ls /data/simulation/outputs/{short_name}/

Find the most recent graph_round*.json and signals_round*.json files (highest round number). If no files are found, stop and report:

✗ nova-economy-contributor: no simulation output found in /data/simulation/outputs/{short_name}/
  The simulation must be run on the host before this skill can contribute.

Read the graph file using the shell tool:

cat /data/simulation/outputs/{short_name}/graph_round{NNNN}.json

Store the full JSON string as file_content. Parse these fields from it:

• graph.round → round number
• graph.host_agent_id → confirm it matches NOVA_ACCOUNT_ID
• graph.n_trade_events → for the report

Read the signals file using the shell tool:

cat /data/simulation/outputs/{short_name}/signals_round{NNNN}.json

Parse the macro block and store these values for the final report:

• market_efficiency
• cooperation_index
• wealth_gini
• strategy_convergence
• defection_count
• volatility_detected

Step 4 — Upload graph to NOVA

Call the nova-submit tool with these exact parameters:

• account_id: value of NOVA_ACCOUNT_ID
• api_key: value of NOVA_API_KEY
• group_id: value of NOVA_GROUP_ID
• filename: {NOVA_ACCOUNT_ID}-round{round}.json
• file_content: the full JSON string read in Step 3

On success, nova-submit returns a cid. Store it as nova_cid.

On failure, set nova_cid = "PENDING" and note the error. Continue to Step 5.

Step 5 — Update memory

Call memory_write with path swarm/last_contribution.md and this content:

# Last Contribution

contributed_at_utc: {current UTC timestamp}
round: {round}
cid: {nova_cid}
group_id: {NOVA_GROUP_ID}
agent_id: {NOVA_ACCOUNT_ID}

Step 6 — Report

If all steps succeeded:

✓ nova-economy-contributor v0.2.0
  Round:        {round}
  CID:          {nova_cid}
  Group:        {NOVA_GROUP_ID}
  Agent:        {NOVA_ACCOUNT_ID}
  Trades:       {n_trade_events}
  Mkt Eff:      {market_efficiency}
  Cooperation:  {cooperation_index}
  Gini:         {wealth_gini}
  Convergence:  {strategy_convergence}
  Defections:   {defection_count}
  Volatility:   {volatility_detected}

If any step failed, append an errors section:

  ─── Errors ───
  NOVA upload: {error or "ok"}

---

What this skill does NOT do

• It does not run the simulation — run_simulation.py is executed on the host before triggering this skill.
• It does not push signals to Qdrant — that is handled by push_signals.py on the host after the CID is known.
• It does not wait for user input at any point.
• It does not retry failed uploads — failed CIDs are recorded as PENDING in memory.