agentleFS
Arcanea-LabsArcanea-Labs/Starlight-Intelligence-System/skills/foundry/system-forge/SKILL.md

system-forge

Skill10 starsChanged 2 months agoMITView on GitHub

Design a multi-capability system as a swarm, vertical, or plugin; compile it only when a Foundry runtime is available. Use when several skills or agents need explicit topology, shared state, conflict resolution, termination, deployment adapters, and cross-surface packaging.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 8 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not create a swarm merely to make a workflow sound advanced.

    line 19
  2. Distinctive:

    Start with a Task Envelope.

    line 23
  3. Distinctive:

    Prefer registered skills and agents.

    line 26
  4. Distinctive:

    Treat lexical graph suggestions as discovery hints, never silent routing decisions.

    line 27
  5. Distinctive:

    Write success conditions, stop conditions, maximum rounds, cost or latency ceilings, and escalation points.

    line 39
  6. Distinctive:

    Keep builder, critic, and verifier separate when stakes or ambiguity justify it.

    line 40
  7. Very specific:

    Foundry runtime available: only when tools/foundry/cli.mjs and foundry/contracts/ exist in the current workspace, compile the appropriate …

    line 42
  8. Distinctive:

    Run behavioral and economic lanes before increasing autonomy or fan-out.

    line 55
  9. Distinctive:

    Every role owns a distinct output or policy boundary.

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

System Forge

Outcome

Design a bounded system whose topology, authority, shared state, and proof obligations are visible before execution, then compile it when a Foundry runtime is present.

Choose the smallest system shape

  • Skill composition: one actor can perform the work sequentially.
  • Swarm: independent or specialized workers need explicit coordination.
  • Vertical: domain rules, audiences, taste, and evaluation recur across several capabilities.
  • Plugin: a set of validated skills and optional MCP connectivity must be distributed to ChatGPT and Codex.

Do not create a swarm merely to make a workflow sound advanced. Parallelism is justified only for genuinely separable work, independent evaluation, latency reduction, or permission isolation.

Procedure

  1. Start with a Task Envelope.
    • Include deliverables, stakes, autonomy, capability selection, permissions, required evidence lanes, and deployment targets.
  2. Resolve capabilities.
    • Prefer registered skills and agents.
    • Treat lexical graph suggestions as discovery hints, never silent routing decisions.
  3. Choose topology.
    • manager-workers: one owner decomposes and synthesizes.
    • handoff: ownership moves between specialists.
    • agents-as-tools: one agent retains control and invokes bounded specialists.
    • parallel-merge: separable outputs join under a deterministic merge.
    • debate-synthesize: independent hypotheses are judged and synthesized.
  4. Define authority and state.
    • Give every role non-overlapping decision rights.
    • Choose manager-only, append-only, or partitioned shared-state writes.
    • Name the conflict owner and evidence-based resolution method.
  5. Define termination before dispatch.
    • Write success conditions, stop conditions, maximum rounds, cost or latency ceilings, and escalation points.
  6. Keep builder, critic, and verifier separate when stakes or ambiguity justify it.
  7. Select the execution mode.

    • Foundry runtime available: only when tools/foundry/cli.mjs and foundry/contracts/ exist in the current workspace, compile the appropriate pack:
    node tools/foundry/cli.mjs forge \
      --envelope <task-envelope.json> \
      --pack <swarm-pack.json|vertical-pack.json|plugin-pack.json> \
      --out <package-directory>
    • Portable skills-only install: read references/portable-contracts.md, then author the Task Envelope and chosen pack with those field names. Mark compilation and proof pending-runtime; do not invent emitted host artifacts or an Evidence Receipt. Provide the command above as an optional handoff for a later SIS workspace run.
  8. Prove static structure before live orchestration.
    • Validate topology, capabilities, permissions, state policy, termination, and plugin shape.
    • Run behavioral and economic lanes before increasing autonomy or fan-out.

Plugin rule

A skills-only plugin is a valid first release. Add an MCP server only when a real deployed endpoint, authentication model, privacy contract, and tool-level tests exist. Never invent a future endpoint in the manifest.

Quality gates

  • Every role owns a distinct output or policy boundary.
  • The synthesizer is named.
  • Conflict resolution is explicit.
  • Shared-state writes cannot silently overwrite another role.
  • Termination and budget ceilings are machine-readable.
  • Deployment claims match surfaces that were actually packaged and tested.

Return

Return the system-shape decision, topology, capability resolution, pack, execution mode, and the smallest blocked next step. Include compiled artifacts and an Evidence Receipt only when they were actually produced.

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.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.