agentleFS
HazATHazAT/pi-config/skills/superconductor/SKILL.md

superconductor

Skill451 starsChanged 7 months agoMITView on GitHub

Operates the super.engineering/Superconductor `sc` CLI for app-managed agents, teams, sessions, layouts, worktrees, and review threads. Use when a workflow explicitly calls for SC orchestration or app-managed UI/session behavior.

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 9 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use live sc instructions and capability output as authoritative.

    line 9
  2. Distinctive:

    Use SC only for the requested orchestration or app-managed outcome.

    line 13
  3. Distinctive:

    Never create a worktree merely for delegation or launch replacement agents for follow-ups.

    line 15
  4. Very specific:

    Prefer stable labels or stable target IDs and --active keep.

    line 16
  5. Distinctive:

    Before the first mutation, run the matching guide:

    line 21
  6. Distinctive:

    Verify the launch shape, provider, UI mode, model, reasoning level, and structured-read support before specifying them.

    line 40
  7. Distinctive:

    Read references/orchestration.md before launching or controlling sessions.

    line 42
  8. Distinctive:

    Check reads for target/provider errors, incomplete output, and the evidence required by the prompt.

    line 65
  9. Very specific:

    Use sc agent send for clarification in the existing session.

    line 67
  10. Very specific:

    Use sc team run only for genuinely independent fan-out whose roles do not require per-role UI or model overrides:

    line 71
  11. Distinctive:

    Every team prompt must include its task, absolute handover paths, read/write limits, and this completion contract:

    line 82
  12. Distinctive:

    Do not create sidecar directories, repository run stores, membership records, coordination-state mirrors, or custom JSONL state.

    line 102
  13. Very specific:

    Verify response.new_title.

    line 112
  14. Distinctive:

    Do not stop agents, close tabs/views, delete groups, or remove other managed state unless the human requested cleanup.

    line 116

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Superconductor CLI

Use live sc instructions and capability output as authoritative. Keep SC syntax and lifecycle policy here; role skills should reference this skill rather than reproduce commands.

Boundaries

  • Use SC only for the requested orchestration or app-managed outcome.
  • Require explicit human intent for worktree/branch creation or deletion, target-branch changes, force termination, destructive cleanup, closing/rearranging existing sessions, and unrelated review-thread mutations.
  • Never create a worktree merely for delegation or launch replacement agents for follow-ups.
  • Prefer stable labels or stable target IDs and --active keep.
  • Dispatch and idle state are not completion; always wait, read, and verify.

Live Preflight

Before the first mutation, run the matching guide:

AreaCommand
Agents, teams, delegationsc instructions orchestration
Tabs, panes, viewssc instructions layout
Worktrees or branchessc instructions worktree
In-app review threadssc instructions review

For orchestration, inspect current capabilities and targets:

command -v sc
sc layout capabilities --output json
sc layout views --worktree "$PWD" --output json
sc agents list --worktree "$PWD" --output json
sc chat providers --json

Verify the launch shape, provider, UI mode, model, reasoning level, and structured-read support before specifying them. If unavailable, report the exact limitation instead of silently substituting another mechanism.

Read references/orchestration.md before launching or controlling sessions. Read references/worktrees-and-reviews.md before requested worktree or review-thread operations.

Individual Pi Worker

Delegated Pi sessions default to terminal mode. Long prompts should use a temporary JSONL launch input built with jq; it is command input, not durable workflow state, and must be removed after launch.

sc layout run tabs \
  --provider pi --ui terminal \
  --label worker-TODO-001 \
  --prompt 'Complete prompt with task and absolute handover paths.' \
  --worktree "$PWD" --active keep --output json

Capture the returned label/stable target. Then collect that same target:

sc agent wait --to label:worker-TODO-001 --idle --timeout-ms 120000 \
  --worktree "$PWD" --output json
sc agent read --to label:worker-TODO-001 --last 40 \
  --worktree "$PWD" --output json

Check reads for target/provider errors, incomplete output, and the evidence required by the prompt. Verify source-writing results independently with Git and tests. Run shared-checkout source writers sequentially.

Use sc agent send for clarification in the existing session. If a turn must be redirected, interrupt then send. Do not infer success from launch, queue admission, or idle state.

Independent Teams

Use sc team run only for genuinely independent fan-out whose roles do not require per-role UI or model overrides:

sc team run \
  --label correctness --provider pi --prompt 'Independent read-only review prompt.' \
  --label regressions --provider pi --prompt 'Independent read-only review prompt.' \
  --worktree "$PWD" --output json

sc team run currently has no per-role --ui, --model, or --reasoning flags. When an ensemble requires distinct Pi models or Pi Chat UI, launch individually labeled sessions with sc layout run tabs, wait/read their stable targets, and fan their reports into the designated lead session. Do not claim those sessions are an SC team run or instruct them to call sc team report.

Every team prompt must include its task, absolute handover paths, read/write limits, and this completion contract:

sc team report --run RUN_ID --role ROLE --status done \
  --summary 'Concise evidence-based report.' --worktree "$PWD" --output json

Use --result-file only when a detailed report exceeds the summary limit; do not use it for the session-file workflow. In a synchronous workflow, capture the returned stable role targets, wait for them, then collect the durable reports with sc team status. Once every role is Reported, use those report summaries directly; call sc agent read only for a missing, failed, or malformed report.

Do not pass --notify self when the coordinator waits synchronously. That option inserts the team synthesis into the creator's editor as a follow-up prompt and may require the user to submit it. Use it only for an explicitly asynchronous workflow, end the coordinator turn after launch, and explain that behavior to the user. Never combine it with manual wait/status collection. Team launch provides parallelism, not sequencing.

Session-File Handover

For /plan, /todos, /execute, and /review:

  • A coordinator session file is required: PI_SESSION_FILE in Pi, or $HOME/.claude/projects/<project slug>/$CLAUDE_CODE_SESSION_ID.jsonl in Claude Code (slug = the project directory path with / and . mapped to -).
  • Durable handover files are exactly ${SESSION_FILE%.jsonl}.plan.md, ${SESSION_FILE%.jsonl}.todos.md, and ${SESSION_FILE%.jsonl}.review.md.
  • /review may use the session JSONL and Git state without plan or todos files; include either handover when it exists.
  • The coordinator alone writes these files.
  • Delegated prompts receive the resolved absolute paths and read them directly.
  • Do not create sidecar directories, repository run stores, membership records, coordination-state mirrors, or custom JSONL state.

Managed Terminal Title

In a managed Pi terminal:

sc tab title "$TITLE" --to "id:terminal:$SUPERCONDUCTOR_TERMINAL_ID" --json

Verify response.new_title. Do not omit the stable target.

Cleanup

Do not stop agents, close tabs/views, delete groups, or remove other managed state unless the human requested cleanup. Inspect exact targets and current help before any such operation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.