agentleFS
asta-nguyenasta-nguyen/agent-devkit/skills/using-devkit/SKILL.md

using-devkit

Skill4 starsChanged 2 months agoMITView on GitHub

Use when starting work in a repository with agent-devkit installed, or when a request does not clearly map to a workflow skill.

  • Deletes or force-pushes

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    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:

    Never put process artifacts in or link them from docs/llm/.

    line 18
  2. Very specific:

    Commit/push only on explicit request, after a final review-and-verify pass.

    line 19
  3. Distinctive:

    Call the next skill directly.

    line 27
  4. Very specific:

    Stop where the user limited the request ("plan only").

    line 34
  5. Distinctive:

    Baseline: before the first source edit, run existing relevant checks and record failures.

    line 36
  6. Distinctive:

    Every turn that ends a skill, stops for the user, or hands off ends with:

    line 42
  7. Very specific:

    Complete: after final review passes, review-and-verify moves the folder to changes/archive/<same name>/ and relinks under ## Archived …

    line 62
  8. Distinctive:

    Not completion: never archive, mark tasks done, or discard code/Git changes.

    line 67
  9. Distinctive:

    No folder: create none; say so.

    line 68
  10. Very specific:

    Delta: create only for an existing wiki requirement needing ADDED/MODIFIED/REMOVED.

    line 71

≈ “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

Using Devkit

Choose the owning workflow skill and follow its instructions. Read context first: read-codebase-context, or setup-codebase on a first visit.

Shared rules

  • Name a new process artifact YYYY-MM-DD-<issue-id>-<slug> (before any type suffix) only when the task or a related artifact, including its filename, gives an issue ID; never infer one or rename an artifact.
  • Link internally with relative Markdown links; convert legacy [[...]] links in any document you edit and report unresolvable targets.
  • Never put process artifacts in or link them from docs/llm/.
  • Commit/push only on explicit request, after a final review-and-verify pass. One branch/PR per task; no locks or coordination tools. For conflicts in docs/agent-devkit/INDEX.md or docs/llm/INDEX.md, preserve every task link and verify targets.
  • Working-tree changes are user-owned: "clean up" never authorizes git reset --hard, git clean, git stash, or deletes, even after a backup. Ask first.

Handoffs and stops

  • Call the next skill directly. Stop only for approval, user question/action, or a stalled fix loop. Route handoffs advance; repair handoffs return to caller.
  • Each route has one approval: Bounded chat design or Architectural design.md + tasks.md (+ delta.md). It authorizes implementation through review; ask again only for a material change to scope, behavior, interface, schema, or risk. An instruction before the approved files existed is not approval.
  • Stop where the user limited the request ("plan only"); never call implement-task without a request to implement.
  • Baseline: before the first source edit, run existing relevant checks and record failures. If none, note it; no baseline-only test/runner. Implementation includes documentation: call document-wiki's ## Task updates when an existing wiki needs an in-scope update; stale or missing coverage blocks completion. Merge, archive, and relinking need no extra approval.
  • Every turn that ends a skill, stops for the user, or hands off ends with:

    Next: <skill or action, or none>
    Waiting on: agent | user — <exact approval, answer, or action needed>
    Done when: <observable result that completes the next step>

    review-and-verify keeps its own result block; its first three lines match.

Change lifecycle

docs/agent-devkit/changes/<name>/ exists only with a process file (design.md, tasks.md, delta.md, estimate.md, decisions.md, or handoff.md); Architectural changes need design.md and tasks.md. On creation (Architectural: on joint approval), link it under ## Changes in docs/agent-devkit/INDEX.md, adding the heading if missing.

  • Open: edit in place; tasks.md's ## Approval Gate records approval, not completion. Material changes reopen it.
  • Complete: after final review passes, review-and-verify moves the folder to changes/archive/<same name>/ and relinks under ## Archived changes.
  • Cancelled: needs the user's reason (ask if missing). Move the folder to changes/cancelled/<same name>/ via review-and-verify's move-and-relink step (links only), listing it under ## Cancelled changes with the reason. Not completion: never archive, mark tasks done, or discard code/Git changes. No folder: create none; say so.
  • Archived and cancelled folders are history: add no requirements, tasks, or results; follow-up starts a new change.
  • Delta: create only for an existing wiki requirement needing ADDED/MODIFIED/REMOVED; not for a domain with no page, missing wiki, or restored behavior. Target ## Requirements pages; convert a legacy ## Business rules page with document-wiki first.
  • Older plans, specs, decision files, and Required/Execution gates are legacy: see references/legacy.md.

Task flows

Classify the task, state the class (the user may override), then follow its row. When in doubt take the heavier row; hidden complexity moves up, never down.

TaskFlow and owning skills
Spike or bug question ("can we…?", "is this a bug?")Investigate → report evidence; no production code: brainstorm-feature (Spike), systematic-debugging (bug question)
Clear, small change (conditions below)Impact check → code → review: brainstorm-feature → implement-task → review-and-verify
Bounded feature (any other change to an existing flow)Settle decisions → chat design → approve → code → review: brainstorm-feature → implement-task → review-and-verify
Bounded bugReproduce → root cause → failing check → fix → passing check → review: systematic-debugging → review-and-verify
Architectural feature or bugClarify → design.md + tasks.md → one joint approval → code → review: brainstorm-feature → plan-feature → implement-task → review-and-verify
  • A clear, small change meets all four: the user named the exact change and file or symbol with an unambiguous result; the impact check keeps every caller, test, config, and wiki page in scope; it touches no public API, schema, dependency, CI, security boundary, or data-loss risk; it is not a bug. brainstorm-feature confirms; if any fails, use the Bounded feature row.
  • A bounded bug changes no shared interface, contract, or component boundary; when its class is unclear, start with systematic-debugging, which moves architectural bugs up.

Other routes: plan-feature revises an open tasks.md or legacy plan (also for a material decision mid-implementation); brainstorm-feature revises an open Bounded delta.md; implement-task resumes an approved change.

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.