Skip to content
Agentmart
Submit a tool

code

Code-workflow scaffolding in one mode-based skill — BRANCH (git branch/worktree isolation with exp|feat|fix prefixes), ROADMAP (step-by-step implementation…

Skill by Shiien

IndexedMIT★ 7
Write a review

Chalk Labs

Need help deploying code?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/shiien-code

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Code-workflow scaffolding in one mode-based skill — BRANCH (git branch/worktree isolation with exp|feat|fix prefixes), ROADMAP (step-by-step implementation roadmap saved to docs/implement_roadmap/ that code-implement executes), DEBUG (rigid four-phase Reproduce → Isolate → Root Cause → Verify, with the three-failed-fixes-means-architecture rule), COMMIT (pre-commit verification, prefixed message, explicit staging, checklist SHA annotation). Absorbs the former code-branch / code-roadmap / code-debug / code-commit skills — those names now refer to modes here. Implementation and review stay separate skills (`code-implement`, `code-review`; codex-track variants). Triggers on "start a branch", "create a worktree", "write the implementation roadmap", "debug this", "test fails", "commit these changes".

Use cases: testing, writing, productivity, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/code

What's inside

Skill name
code
Description
Code-workflow scaffolding in one mode-based skill — BRANCH (git branch/worktree isolation with exp|feat|fix prefixes), ROADMAP (step-by-step implementation roadmap saved to docs/implement_roadmap/ that code-implement executes), DEBUG (rigid four-phase Reproduce → Isolate → Root Cause → Verify, with the three-failed-fixes-means-architecture rule), COMMIT (pre-commit verification, prefixed message, explicit staging, checklist SHA annotation). Absorbs the former code-branch / code-roadmap / code-debug / code-commit skills — those names now refer to modes here. Implementation and review stay separate skills (`code-implement`, `code-review`; codex-track variants). Triggers on "start a branch", "create a worktree", "write the implementation roadmap", "debug this", "test fails", "commit these changes".
License
MIT
Instructions
85 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# code

Writing the code is `code-implement`; reviewing it is `code-review` (both
ship codex-track variants). Chain: BRANCH → ROADMAP → `code-implement` →
DEBUG (as needed) → `code-review` → COMMIT.

**Shared context**: Read `skills/_shared/git-conventions.md` for branch
naming, worktree pattern, commit prefixes, staging rules, and the
never-commit list — the modes below don't restate them.

## Mode: BRANCH — "start a branch", "isolate this work"

| Signal | Path |
|---|---|
| Small edit (<30 lines, single file, <1h) | stay on current branch — report and skip |
| New feature / multi-file change | `git checkout -b {exp|feat|fix}/{slug}` |
| Long-running or must-not-disturb work | `git worktree add .worktrees/{type}-{slug} -b {type}/{slug}` then work there |
| Experiment sweep / ablation / probe | named `exp/*` branch |

Report the branch/worktree to the user and to downstream modes.

## Mode: ROADMAP — medium/large task needs a plan before code

1. Understand the goal (ask if ambiguous), read the relevant sources and
   `methodology/approach.md`, identify the test framework + exact test
   command.
2. Save `docs/implement_roadmap/YYYY-MM-DD-{name}.md` — **every section
   mandatory** (this format is the execution contract `code-implement`
   Track A/B both consume):
   - `## Goal` — one sentence, end state.
   - `## Context` — project, language/framework, **test command verbatim**,
     key files with one-line whys.
   - `## Constraints` — never modify `.claude/`, `skills/`, `memory/`,
     `hooks/`, `config.yaml`, `CLAUDE.md`; no `git commit/push` during
     execution; no new deps unless justified in a step; + project-specific.
   - `## Steps` (3-8; fewer → maybe no roadmap needed, more → split): each
     step has **Purpose / Files (Create·Modify·Test) / What to do**
     (function-class level, describe not code) **/ Acceptance** (runnable
     command or verifiable condition).
   - `## Done Criteria` — full test suite + functional check + regression
     check.
   - `## Status` — one checkbox per step, base commit, last updated.
3. Quality check before presenting: acceptance runnable, key files exist,
   constraints include framework dirs, test command verbatim.
4. Driven by a checklist Implementation checkbox → append the roadmap path
   as its artifact (`checklist` update mode).

## Mode: DEBUG — any bug, test failure, unexpected behavior (RIGID phases)

1. **Reproduce** — run the exact failing command yourself; never trust
   second-hand descriptions; establish the failure rate if flaky.
2. **Isolate** — binary-search the smallest failing unit; check
   `git diff` / `git log` (bisect if it used to work); targeted assertions,
   not print-spray.
3. **Root cause** — the cause, not the symptom. **Hard rule: three failed
   fix attempts in a row → STOP**, report the three hypotheses and why each
   failed, and raise "this is architectural, not implementation" to the
   user. Do not attempt fix 4.
4. **Verify** — fix addresses the root cause; original failing command
   passes; full suite passes; temporary debug code removed. → `code-review`
   → COMMIT with `fix:`.

## Mode: COMMIT — after `code-review` passes

1. **Pre-commit verification**: full test command (from the roadmap
   Context) passes; each Done Criterion checked with evidence;
   `git diff --stat` presented; explicit user confirmation ("Proceed?
   y/n" — never commit on ambiguity). Small no-roadmap tasks keep the diff
   summary + confirmation only.
2. Prefix by dominant intent: `feat: | fix: | exp: | refactor: | docs: |
   chore:` — imperative subject ≤72 chars; body explains WHY.
3. Stage **explicitly by filename** (never `-A`); `git status` against the
   never-commit list.
4. Post-commit: tree clean, `git log -1` sane, capture the SHA.
5. Checklist-driven tasks → `checklist` (update mode): mark `[x]` and
   annotate the checkbox with the roadmap path + short SHA. Push is
   separate, per git-conventions.

**Inputs**: task intent / failing output / reviewed tree
**Outputs**: branch·worktree / `docs/implement_roadmap/*.md` / fixed code / commit
**Token**: BRANCH 1-2K · ROADMAP 3-8K · DEBUG 3-15K · COMMIT 2-5K
**Composition**: ROADMAP → `code-implement`; DEBUG fix needing real logic →
`code-implement` TDD loop then back to phase 4; COMMIT ← `code-review` PASS.

From Shiien/Self-Evolved-Research-Framework, MIT.

More from Shiien/Self-Evolved-Research-Framework

Signals

Stars
7
Forks
1
Open issues
0
Last commit (this folder)
Jul 13, 2026 (3 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from Shiien

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get code running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying code?

CallGet help
Cite this page

code skill. Agent Mart. https://agentmart.dev/skills/shiien-code (last updated 2026-10-24)