fyrlove/r0re/r0re-blackboard

r0re Blackboard

You are one of several workers analyzing the same target. The blackboard is the single source of truth shared by all workers: facts everyone can use, dead ends nobody should retry, intents waiting to be claimed.

The client is blackboard.py (pure stdlib, already on PATH or at /opt/r0re/skills/r0re-blackboard/blackboard.py). Environment variables R0RE_BLACKBOARD_URL, R0RE_PROJECT_ID, R0RE_WORKER_NAME are pre-injected — never invent them yourself.

Discipline — when to read

Run python3 /opt/r0re/skills/r0re-blackboard/blackboard.py read (and read-deadends) at these moments:

1. Before starting any new analysis direction — do not repeat what teammates

already confirmed, and never retry a recorded dead end.

1. When stuck — scan read for facts that unblock you.
2. Before claiming work — list-intents shows what is available.

Discipline — when to write

1. The moment you confirm a finding in real tool output, write it:

write-fact "<concise finding with exact location>" --verified Only pass --verified for things you literally saw in command output.

1. The moment a route is disproven, record it:

mark-deadend "<what was tried and why it failed>" A dead end is as valuable as a fact — it stops everyone from repeating it.

1. Keep facts atomic and objective: one finding per write, with file path /

class / offset / symbol. Do not write speculation as verified.

Discipline — intents

1. list-intents → pick one → claim <id>. If it prints LOST, claim a

different one. Never work on an intent you did not win.

1. When done with a claimed intent, always conclude <id> "<result summary>",

even if the outcome is "nothing found" — that closes the loop for everyone.

1. Only call complete --from <fact-ids> "<conclusion>" when the project goal

is genuinely achieved and backed by facts you wrote. The server gate will reject unsupported completions.

Rules

• The blackboard is append-only: never try to edit or delete another worker's

records. If a fact looks wrong, write a new fact with the corrected evidence.

• Write in the same concise, location-precise style as the facts you read.
• Large outputs belong in files under the analysis workspace; the fact should

reference the path.