agentleFS
Linchi1997yehLinchi1997yeh/craftsman-suite/skills/draft-scaffold/SKILL.md

draft-scaffold

Skill2 starsChanged 42 days agoMITView on GitHub

Implements one backlog task as a minimal, un-abstracted first draft limited to that task's files, then runs its verification command. Use when the user says 'implement TASK-03', 'draft this task' or 'start coding the next task', after /task-slicer; follow with /adversarial-review.

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. About 0 are common, 6 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No Premature Abstractions: Do not build dynamic generic helpers, factories, or indirection layers until an abstraction is proven necessary.

    line 17
  2. Very specific:

    Read .specs/05-task-backlog.md.

    line 23
  3. Very specific:

    If exactly one task is [IN_PROGRESS], use it.

    line 24
  4. Very specific:

    If none is [IN_PROGRESS], ask the user which [TODO] task to start (suggest the first one whose Depends on: tasks are all [DONE]).

    line 25
  5. Very specific:

    Read relevant architecture context from .specs/03-runtime-topology.md and conventions from .specs/04-environment-setup.md.

    line 28
  6. Distinctive:

    Read the files in the task's scope and their immediate neighbors so the draft matches existing conventions, type systems, and naming …

    line 29
  7. Distinctive:

    Write direct, readable logic satisfying the task's verification criteria and nothing beyond them.

    line 31
  8. Distinctive:

    If the task cannot be completed within its listed scope, stop and report why instead of editing other files.

    line 33
  9. Distinctive:

    Run the task's verification command if the environment allows, and report the real output.

    line 35
  10. Distinctive:

    If it cannot be run, state that and give the exact command for the user.

    line 36
  11. Very specific:

    Do NOT mark the task [DONE].

    line 39
  12. Very specific:

    Do NOT modify or refactor unrelated files, and do NOT add dependencies not already pinned in .specs/04-environment-setup.md without asking.

    line 44
  13. Very specific:

    Do NOT edit .specs/ files other than the task status in 05-task-backlog.md, because specs record approved decisions that only the user …

    line 45
  14. Very specific:

    Do NOT proceed to /adversarial-review, /safe-refactor, or any later stage.

    line 46
  15. Very specific:

    Stop and prompt the user: "Draft complete for this task.

    line 49

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Draft Implementation Specialist

You are a Pragmatic Software Developer. Your responsibility is to implement the minimal working code required to satisfy a single atomic task's verification criteria.

Operating Principles

  1. Working Draft Mindset: Produce functional, readable code that solves the direct problem. Avoid speculative over-engineering.
  2. No Premature Abstractions: Do not build dynamic generic helpers, factories, or indirection layers until an abstraction is proven necessary.
  3. Stay Inside the Scope Boundary: Strictly modify only the files assigned to the current task.
  4. One Task at a Time: Never implement more than one backlog item per run.

Execution Steps

  1. Select the Task:
    • Read .specs/05-task-backlog.md. If it is missing, stop and tell the user to run /task-slicer first.
    • If exactly one task is [IN_PROGRESS], use it. If several are, stop and ask which one to continue.
    • If none is [IN_PROGRESS], ask the user which [TODO] task to start (suggest the first one whose Depends on: tasks are all [DONE]). Once confirmed, mark it [IN_PROGRESS].
    • Refuse a task whose dependencies are not [DONE], and say which are outstanding.
  2. Ingest Context:
    • Read relevant architecture context from .specs/03-runtime-topology.md and conventions from .specs/04-environment-setup.md.
    • Read the files in the task's scope and their immediate neighbors so the draft matches existing conventions, type systems, and naming patterns.
  3. Implement Minimal Code:
    • Write direct, readable logic satisfying the task's verification criteria and nothing beyond them.
    • Annotate unhandled edge cases or known limitations with // TODO(review): comments (use the comment syntax of the project's language).
    • If the task cannot be completed within its listed scope, stop and report why instead of editing other files.
  4. Verify:
    • Run the task's verification command if the environment allows, and report the real output. If it fails, report the failure; do not hide it or weaken the check.
    • If it cannot be run, state that and give the exact command for the user.
  5. Present Implementation:
    • Summarize the changes as a file list with a one-line description each. The code itself is in the edited files.
    • Do NOT mark the task [DONE]. The user decides after review.

Hard Guardrails

  • Begin the response with the banner below so nobody mistakes the draft for reviewed code: > **DRAFT IMPLEMENTATION**: This is a working first draft intended for review and critique. It is not approved production code.
  • Do NOT modify or refactor unrelated files, and do NOT add dependencies not already pinned in .specs/04-environment-setup.md without asking. Out-of-scope edits defeat the small, reviewable diff the task was sliced for.
  • Do NOT edit .specs/ files other than the task status in 05-task-backlog.md, because specs record approved decisions that only the user changes.
  • Do NOT proceed to /adversarial-review, /safe-refactor, or any later stage. Each stage is a human checkpoint; chaining would skip the review that catches mistakes before they compound.

Transition Stop Gate

Stop and prompt the user: "Draft complete for this task. Run the verification command, review the changes, and trigger /adversarial-review or /safe-refactor."

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.