agentleFS
chipweaverchipweaver/veripower/skills/design-flow/SKILL.md

design-flow

Skill10 starsChanged 53 days agoMITView on GitHub

Use when progressing IC design through stages, checking module status, or routing rework decisions; not for single-stage execution or artifact authoring.

9 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    Use kernel.py status --module {module} to check current status.

    line 18
  2. Distinctive:

    Run the kernel loop:

    line 23
  3. Very specific:

    After launching a background executor, query decide again so other ready work can proceed.

    line 42
  4. Distinctive:

    Follow the user's actual authorization.

    line 60
  5. Very specific:

    When the task calls for a recorded acceptance, use decide --closing to check the current conclusions and delivered artifacts.

    line 75

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “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

Design Flow

Coordinate module stages, rework and delivery through python3 <skill>/../../framework/scripts/kernel.py, written kernel.py below. <skill> is this skill's directory; {module} contains events.jsonl, intent/brainstorm.md and the stage trees.

The kernel owns events and scheduling. Stage executors own their workdir artifacts and close through the stage CLI, whether running in this conversation or in a subagent. While orchestrating, use the kernel rather than editing events or stage results directly. Inspect source when needed to understand an operation.

Use kernel.py status --module {module} to check current status. Command arguments are available through kernel.py <command> --help.

Run the work

Run the kernel loop:

loop:
  a = kernel.py decide --module {module} [--wake <rule>:<run>] [--closing]
  execute(a)  # actions below; launching a background executor does not wait for it
  if a.action == DONE: finish

For DISPATCH, invoke the returned dispatch_args through the kernel CLI. The response prepares the workdir and identifies the executor; it does not start that executor.

executionStart the work
main-threadInvoke the returned skill in this conversation, acting as that stage's executor.
taskRender the stage prompt with the module, rule, skill and workdir from dispatch; launch an independent background executor with that prompt through the host adapter.

Skill names are bare stage names. The host adapter supplies native skill loading and delegation.

After launching a background executor, query decide again so other ready work can proceed. Confirm executor exit through the host before waking or reaping its run; a result file or YIELD does not establish exit. Once it exits, pass --wake <rule>:<run> to decide, including when it left no result. Carry out the returned action:

  • REAP: run kernel.py reap --module {module} --rule <rule> --run <run>, then query again. Reap uses that run's result; an absent result is handled as an incomplete execution.
  • YIELD: report the running work and wait for an executor completion before resuming.
  • ESCALATE: address the reported blocker under the authorization below, then query decide again. Wait when needed information or a reserved human decision is unavailable. kernel.py diagnose records a later addition or correction to repair ownership; --supersedes identifies the diagnosis being replaced.
  • DONE: deliver the conclusions and complete any acceptance in scope, as described below.

A failed CLI operation remains incomplete until its reported cause is resolved.

Decisions and authorization

Follow the user's actual authorization. For a reserved decision, present the specific choice, evidence and unresolved issues, then wait. Within explicit delegation, decide and act in scope. A refusal remains binding. Host execution permission is separate from decision authorization. For diagnose and signoff, --provenance identifies the decision maker and authorization; --reason gives the engineering grounds.

For artifact changes, kernel.py consequences --module {module} --paths <path…> shows which valid proofs the change would invalidate.

Completion and acceptance

Report what the delivered evidence establishes and any remaining limits. Recheck concrete issues through the responsible stage and existing repair flow. Completion requires separate approval only when the task reserves that decision.

When the task calls for a recorded acceptance, use decide --closing to check the current conclusions and delivered artifacts. Present the returned basis, follow the authorization above, and record acceptance with:

kernel.py signoff --module {module} --provenance "<decision maker and authorization>" \
  --reason "<basis for accepting this delivery>"

The record applies to the accepted evidence. New stage conclusions need a new acceptance when that is in scope; repeating collection of unchanged evidence does not. An acceptance record does not establish a missing technical fact or resolve a failed check.

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.