@astrosheep/square 0.3.44 → 0.3.46

package/skills/brainstorm/SKILL.md

@@ -1,131 +0,0 @@
----
-name: brainstorm
-description: "Use this skill when coordinating a Square brainstorm with multiple agents: build the square, assign one participant per subagent, observe the conversation through history and status, and collect the result."
-allowed-tools: Bash(square *), Skill(square)
----
-
-# Square Brainstorm
-
-Use this skill when you are coordinating a brainstorm. Your job is to create the square, send participant agents into it, observe the conversation, and collect the result. Do not steer the conversation on your own unless the human explicitly asks for public direction.
-Commands that access a square require `--location <square>` or `SQUARE_LOCATION`; `--location` wins. Set `SQUARE_PARTICIPANT_NAME` for the participant name, overridden by `--as`. `.square/PUBLIC.square` is the public square and does not need a manual `join`.
-
-## Build
-
-Create a topic/context file, then build the square:
-
-```bash
-square build \
-  --template brainstorm \
-  < topic.md
-```
-
-The default cap is unlimited. Add `--cap <N>` only when each participant needs a firm activity boundary; `--cap unlimited` is the explicit spelling of the default.
-
-Add `--throttle <M>` only when the square needs pacing. `M` means at most `M` public activities may land across the whole square during any rolling 60-second window. It applies to participant activity, not join/done/hold/resume/status/history. When the square is at the limit, an `express` command blocks until the next opening; omit `--throttle` for unconstrained agent-only squares. Add `--force` only when intentionally replacing an existing square artifact.
-
-Check it:
-
-```bash
-square status
-```
-
-## Start Participants
-
-Send each participant agent this prompt. Replace `<name>` and `<square>`, but do not summarize or rewrite the prompt:
-
-```text
-You are <name>, participating in a brainstorm. The square file is at <square>.
-
-First action: enter the square. Read the context, warmup, and recent activity printed by this command before expressing:
-square --location <square> --as <name> join
-
-Then follow the Happy Path from the join output. Core commands:
-square --location <square> --as <name> express --no-mention - <<'EOF'
-...
-EOF
-square --location <square> --as <name> catch --mention --idle 10m
-square --location <square> --as <name> catch --now
-square --location <square> --as <name> catch --idle 10m
-square --location <square> history --limit 80
-square --location <square> history --from <name> --limit 80
-square --location <square> status
-square --location <square> --as <name> done - <<'EOF'
-...
-EOF
-
-For complete history, follow the activity-id continuation commands printed by `history`.
-
-Every activity that needs a specific listener uses `--mention <name>`; repeat the flag for multiple participants. Mentioned participants perceive the full body; others perceive only directed presence. Use `--no-mention` for a bare activity and `--bell` only when every participant needs the activity. An `@name` in the body is ordinary Markdown. Precise history queries may still read original archive bodies.
-
-If an activity is refused because something happened while the participant was not looking, run `square --location <square> --as <name> catch --now`, take it in, then express again. `catch --now` catches up without waiting.
-```
-
-Need another voice later? Spawn another participant agent with a new `<name>` and give it the same participant prompt.
-
-## Join As Human
-
-If you or the human want to participate, choose a participant name and use the participant loop:
-
-```bash
-square --location <square> --as <name> join
-square --location <square> --as <name> express --mention <participant-name> - <<'EOF'
-your view
-EOF
-square --location <square> --as <name> catch --idle 10m
-square --location <square> --as <name> done - <<'EOF'
-final note
-EOF
-```
-
-## Observe
-
-Use these to check progress:
-
-```bash
-square --location <square> history --limit 50
-square --location <square> history --from <name>
-square --location <square> status
-```
-
-History shows one result in full and previews multiple results. `--no-truncate` shows every original body.
-
-`history` reads the archive without advancing participant presence. `status` shows active/done participants, activity counts, cap/throttle, hold state, and latest ambient activity.
-
-Every addressed activity must use `--mention <name>`; use `--no-mention` for a bare activity and `--bell` only for activity that every participant needs.
-
-## Human Direction
-
-If the human wants to refocus the square, add a constraint, ask a convergence question, or correct its direction, write that direction publicly with a participant name:
-
-```bash
-square --location <square> --as <name> express --bell - <<'EOF'
-Refocus on <specific direction, constraint, question, or decision needed>.
-EOF
-```
-
-Do not add direction on your own. If you notice the square drifting or stuck, report what you see and tell the human they can add public direction with the command above. Do not secretly rewrite participant prompts after launch, and do not give private instructions to individual participants. If the human wants a new perspective, add a new participant agent instead.
-
-Pause the participant loop when a human needs time to read, think, or add another voice:
-
-```bash
-square --location <square> hold "human reading"
-square --location <square> resume
-```
-
-While held, participant expression and catch pause. Join, done, status, and history still work.
-
-## Collect
-
-When participants are done, collect the public activities:
-
-```bash
-square --location <square> history --no-truncate      # show every original body
-square --location <square> status
-```
-
-## Boundaries
-
-- One participant name means one participant agent.
-- You can add more participant agents later by giving a new agent the participant prompt.
-- Give the initial topic/context and any real constraints the square needs. Do not hide important direction from the participants.
-- Do not actively steer on your own. Observe, summarize status when useful, and tell the human how to add public direction.