# Ywc Sequential Executor

> >-

- **Type:** Skill
- **Install:** `agentstack add skill-yongwoon-ywc-agent-toolkit-ywc-sequential-executor`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [yongwoon](https://agentstack.voostack.com/s/yongwoon)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [yongwoon](https://github.com/yongwoon)
- **Source:** https://github.com/yongwoon/ywc-agent-toolkit/tree/main/claude-code/skills/ywc-sequential-executor

## Install

```sh
agentstack add skill-yongwoon-ywc-agent-toolkit-ywc-sequential-executor
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Sequential Executor

**Announce at start:** "I'm using the ywc-sequential-executor skill to run tasks through their full lifecycle in strict order."

Execute tasks generated by the ywc-task-generator skill through the full development lifecycle: branch creation, implementation, commit, PR, CI verification, merge, and local sync.

## Rationalization Defense

When tempted to skip a step, check this table first:

| Excuse | Reality |
|---|---|
| "Task A and Task B are independent, run them in parallel" | This is the **sequential** executor. For parallel runs, the user must explicitly switch to ywc-parallel-executor. |
| "CI is taking long, --skip-ci-wait gets the next task started" | Skipping CI hides regressions in subsequent tasks. Skip only if user explicitly approved. |
| "Previous task partially succeeded, I'll continue and clean up later" | Each task must leave the codebase buildable. Stop and report on partial success. |
| "Merge conflict is small, I'll resolve and continue" | Conflicts during sequential execution mean a base branch shifted. Stop, sync, get user direction. |
| "PR is created, mark task complete" | Task is complete only when merged + cleaned up + branch deleted. PR alone is not done. |
| "Task spec is unclear, I'll infer from neighbors" | Stop. Inferring from neighbors compounds error across the sequence. Ask user. |
| "Local merge with `--local-merge` is faster, default to it" | Default is PR-based. `--local-merge` only when user explicitly opts in (e.g., personal repo). |
| "Lint/format failed in Step 4 with --draft, I'll stop and report" | Lint/format failures are auto-fixable. Run the project's fix command (e.g., `eslint --fix`, `prettier --write`, `ruff --fix`), commit the result, re-run verification, then continue to Step 5. Only stop if auto-fix fails to resolve residual errors after 2 attempts. |
| "--draft creates the PR and stops — bot review is for later" | Bot review bots (CodeRabbit, Codex Review, Claude Review) post on draft PRs immediately. After finish-branch creates the draft PR, poll for reviews and invoke ywc-handle-pr-reviews. The PR stays draft — responding now avoids a round-trip after un-drafting. |
| "normal-pr range: CI passed and no bot comments yet — proceed to merge" | Bot reviewers post 1–5 minutes after CI completes, not immediately. `ywc-finish-branch` runs a mandatory 60-second initial wait plus a 300-second polling window before concluding no bots are active. `BOT_COUNT == 0` right after CI is not evidence that no bots are active — it means they have not posted yet. The polling window is a required wait gate in range mode, not a pause to skip. |
| "Bot comments were addressed via ywc-handle-pr-reviews — the PR is handled, proceed to merge" | Addressing comments clears only one of three blockers. The review-fix push triggers a fresh CI run that can fail, and the base may have advanced into a `CONFLICTING` state meanwhile. `ywc-finish-branch` re-verifies CI green **and** `mergeable == MERGEABLE` (Step 4 + Step 4-final merge-readiness gate) after every bot iteration — never merge on the strength of "comments addressed" alone. CI failure and base conflict are delivery blockers independent of comments. |
| "Last task in `--local-merge` range: no following task needs this merge, skip finish-branch" | Step 5 (Delivery) is unconditional — **every** task, including the last, must have its feature branch merged into base, the completion-marker committed, and the push executed via `ywc-finish-branch`. Without it, the implementation code is stranded on an orphaned feature branch, `tasks/completed/` is wrong, and the remote base branch is missing the work. The absence of a next task is not a reason to skip delivery. |
| "concurrency/idempotency Task Verify is slow or hard to satisfy locally — pass lint/typecheck/build instead" | Task Verify Layer-1 already gates the merge. A hard-to-satisfy concurrency check must not be downgraded to Layer-4 (lint/build); satisfy it via the task's alternative-verification note (code-level lock/transaction proof or an integration test plan). |
| "Last task in normal-pr range: it's the last one, no need to omit `--defer-push`" | `--defer-push` must be **omitted** on the last task in `normal-pr` range — that omission IS the flush. If `--defer-push` is set on the last task, all accumulated completion-marker commits stay local and are never pushed. The last task's `ywc-finish-branch` invocation performs the single batch push; no separate `git push` runs after the loop. |
| "`--aggregate-pr`: all tasks local-merged onto the work branch, the run is done" | The work branch is not the base branch. The group is delivered only when the single `work → base` PR is created, marked ready, CI-verified, bot-reviewed, **merged**, and local base synced (Section C of [references/aggregate-pr.md](./references/aggregate-pr.md)). A work branch full of merged tasks with no merged PR is an incomplete group. |
| "`--aggregate-pr`: per task, deliver to the real base like local-merge" | Per-task delivery targets the **work branch** (`--base-branch work/` in finish-branch local-merge), never the real base. Merging tasks straight onto the real base would defeat the single-PR goal and mutate base before review. Each task branches from and merges into the work branch only. |
| "`--worktree`: the cycle runs in the worktree, so the skill just `cd`s there once and proceeds" | This skill is an LLM prompt with no persistent shell — a `cd` does not survive between Bash calls. Every git command must be `git -C "$WT" …`, every Edit/Write must target an absolute path under `$WT`, and every test/lint must be `cd "$WT" && ` inside a single Bash call. See [references/worktree-run.md](./references/worktree-run.md) §A1. |
| "`--worktree` non-aggregate DONE: prune the worktree and its integration branch together" | Non-aggregate prune **must** pass `--keep-branch` so the integration branch survives — it is what the user opens the final integration→trunk PR from. Omitting `--keep-branch` deletes that branch and strands the whole run's work with no path to trunk. Only `--aggregate-pr` (whose `work/` is already merged) prunes the branch. See [references/worktree-run.md](./references/worktree-run.md) §A3. |
| "I'll add tests after the implementation is working" | For behavior changes, the failing test comes **first** — a bugfix needs a regression test that fails on the old code, a feature needs a failing unit/integration test before implementation. Without test feedback gating each step you outrun your headlights and the code crashes at runtime. Docs/config/mechanical tasks are the only exception (record it). See [../references/tdd-deep-module-gray-box.md](../references/tdd-deep-module-gray-box.md) §2. |
| "This task only changes internals, no contract to write down" | If the change alters a public contract (exported function, endpoint, event, DTO, schema, props, CLI flag), design and write the interface **before** the body, and report it. Shallow single-use wrappers around cohesive behavior are the maze AI gets lost in later. See [../references/tdd-deep-module-gray-box.md](../references/tdd-deep-module-gray-box.md) §3. |
| "I reviewed the whole implementation line by line" | Default review is gray-box — verify the contract, delegate internals. The exception is **critical paths** (auth, payment, crypto, PII, external input): those get internal review **and** `/ywc-security-audit`, forced even without `--review`. Uniform-depth review wastes effort on safe code and under-scrutinizes dangerous code. See [../references/tdd-deep-module-gray-box.md](../references/tdd-deep-module-gray-box.md) §4. |

**Violating the letter of these rules is violating the spirit.** Sequential execution exists because each task's correctness depends on the previous task's stable state.

## Arguments

Parse `$ARGUMENTS` for the following parameters:

| Parameter | Format | Example | Description |
|-----------|--------|---------|-------------|
| Task specifier | `` or `..` | `000001-010-db-create-users` or `000001-010..000002-030` | Single task or range (phase+sequence prefix match). Both `001010` (legacy) and `000001-010` (new 6-digit PHASE) formats are accepted; prefix matching works with either. |
| `--pr-lang` | `--pr-lang ` | `--pr-lang ja` | PR title/description language. Default: auto-detect from CLAUDE.md or AGENTS.md, fallback to project's dominant language |
| `--tasks-dir` | `--tasks-dir ` | `--tasks-dir ./docs/tasks` | Tasks directory path. Default: `tasks/` |
| `--skip-ci-wait` | flag | | Skip CI wait and auto-merge (create PR only) |
| `--draft` | flag | | Create PR as draft, skip merge |
| `--local-merge` | flag | | Skip PR creation entirely — merge the feature branch into the base branch locally and push. Mutually exclusive with `--draft` and `--skip-ci-wait` |
| `--base-branch` | `--base-branch ` | `--base-branch develop` | Base branch override. Default: auto-detect (develop > main > master) |
| `--dry-run` | flag | | Show the execution plan (task order, dependencies, modes) without executing anything |
| `--terse` | flag | | Compact Completion Report: task table + Completion Status only — no prose reminders, no mode explanations, no advisor notes |
| `--review` | flag | | Auto-run /ywc-impl-review after each task, before PR creation or merge |
| `--run-tests-locally` | flag | | Before merging in `--local-merge` mode, detect and run the project's test command. On failure: mark task FAIL and do not merge. Ignored in PR-based modes. |
| `--aggregate-pr` | flag | | Deliver the whole range as **one work branch + one PR**: each task is local-merged onto a `work/` branch sequentially, then a single PR delivers the group to the base branch (ready → CI → bot → merge). The real base is never mutated until the final merge. See [references/aggregate-pr.md](./references/aggregate-pr.md) |
| `--group-name` | `--group-name ` | `--group-name project-health` | Names the work branch (`work/`). `--aggregate-pr` only; defaults to `work/-` when omitted |
| `--worktree` | flag | | Run the **whole range inside one isolated git worktree** so the main checkout stays free. Run-level isolation — tasks still run strictly sequentially. **Independent flag**, not part of the mutual-exclusion group: combines with all four delivery modes and `--review`. Full lifecycle in [references/worktree-run.md](./references/worktree-run.md). |

**Flag conflicts**: `--local-merge`, `--draft`, `--skip-ci-wait`, and `--aggregate-pr` are mutually exclusive. If the user passes more than one, stop **before any branch or implementation work** and ask which mode they actually want. The reason is that `--local-merge` skips PRs entirely while the others assume a PR exists — silently picking one would surprise the user. `--group-name` is valid only with `--aggregate-pr`. `--worktree` is **orthogonal** to this group — it is not a fifth member and may combine with any one delivery mode (§A8 S4); see [references/worktree-run.md](./references/worktree-run.md).

Example:
```text
/ywc-sequential-executor 000001-010 --local-merge --draft
# → Stop. Report: "--local-merge and --draft are mutually exclusive
#   (local-merge produces no PR; draft requires one). Which mode did you want?"
```

`--review` can be combined with any delivery mode flag.

`--run-tests-locally` has no effect without `--local-merge` (PR CI handles tests in PR-based modes).

If no task specifier is given, detect the next executable task from the dependency graph.

## Dry Run Mode

When `--dry-run` is set, perform Pre-flight and Task Resolution as usual, then display:

- Resolved task list in execution order (with phase+sequence and full name)
- Dependency status for each task (satisfied / pending)
- Active mode flags (`--draft`, `--skip-ci-wait`, `--local-merge`, `--aggregate-pr`, or normal)
- Base branch that would be used (and, for `--aggregate-pr`, the `work/` branch that would be created)

Do **not** create branches, modify files, or run any git commands beyond read-only operations (`git branch --show-current`, `git status`). Exit after displaying the plan.

## Context

- Current branch: !`git branch --show-current`
- Changed files: !`git status --short`
- Tasks directory: !`ls tasks/ 2>/dev/null || echo "No tasks/ directory found"`

## Pre-flight

> **Resume check first**: Before running the checks below, look for `.ywc-run-state.json` in the project root. If it exists, follow the **Resume Detection** procedure in [Checkpoint and Resume](#checkpoint-and-resume). If the user confirms resume, skip Pre-flight and jump to the saved task and step. If the user declines or there is no state file, proceed with the checks below. **Intent-match guard (do not skip)**: if this invocation specifies an explicit range/task that does **not** match the saved run's `range`, never silently resume — surface the divergence and require an explicit choice (resume the saved run vs discard it and start the new range), per the guard in [references/checkpoint-resume.md](./references/checkpoint-resume.md). A new range hijacked by a stale interrupted run is the failure this guard prevents.

Before starting execution, verify these conditions:

1. **Clean working tree** — `git status --porcelain` must be empty. If dirty, warn the user and stop.
2. **On base branch** — Should be on main/develop/master. If on a feature branch, warn and ask whether to continue.
3. **gh CLI ready** — `gh auth status` must succeed.
4. **Tasks directory exists** — The tasks directory must contain task subdirectories and `dependency-graph.md`.
5. **Spec-Reference external URL policy** — Determine whether this project allows fetching external URLs (Notion, Confluence, Figma, etc.) listed in a task's `Spec Reference` section. See [External URL Policy](#external-url-policy) below. This check runs **once per project**, not once per task.

**State Init (non-resume runs only)**: Initialize `.ywc-run-state.json` now using the Write tool (see format in [Checkpoint and Resume](#checkpoint-and-resume)). Do not modify `.gitignore` during execution; the ignore entry must already be committed or present in repo-local exclude before the run:
```bash
grep -qxF '.ywc-run-state.json' .gitignore 2>/dev/null \
  || grep -qxF '.ywc-run-state.json' .git/info/exclude 2>/dev/null \
  || { echo "Missing .ywc-run-state.json ignore entry; add it to tracked .gitignore before the run or to .git/info/exclude for local-only state."; exit 1; }
```

## External URL Policy

Tasks generated by `ywc-task-generator` may list external URLs in `Spec Reference > Primary Sources`. Fetching external content mid-task is unpredictable (network, rate limits, SSO walls) and is also a privacy decision the user must make once per project.

**Procedure**: at Pre-flight, check `.claude/settings.local.json` for `taskExecutor.externalSpecUrls`. If present, use it silently. If missing, ask the user **once** to choose `deny` (default), `allow`, or `allowlist`, then persist the decision under the `taskExecutor` key (preserving every other key in the file). During Step 1b enforcement, apply the chosen policy: `deny` skips all `http(s)://` URLs, `allow` fetches everything, `allowlist` fetches only matching host+path prefixes.

**For full procedure, jsonc schema, exact prompt wording, and rationale**, see [references/external-url-policy.md](./references/external-url-policy.md).

## Definition of Done

A task is **done** only when **all** of the following have happened, in this order. A task that has been "merged but not moved to `completed/`" is **not** done — Step 6's auto-detect logic and other tasks' dependency validation read `/completed/` as the source of truth, so a missing move silently breaks dependency resolution for the next task in any range and for any future invocation of this skill.

1. Implementation passed all four layers of Step 4 verification (Task Verify commands, task-local tests, full project test suite, lint/typecheck/build).
2. `ywc-finish-branch` retur

…

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [yongwoon](https://github.com/yongwoon)
- **Source:** [yongwoon/ywc-agent-toolkit](https://github.com/yongwoon/ywc-agent-toolkit)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-yongwoon-ywc-agent-toolkit-ywc-sequential-executor
- Seller: https://agentstack.voostack.com/s/yongwoon
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.