# Task Generator

> Automatically generates decentralized task files based on manager instructions.

- Skill: `mokhtarabadi/task-generator` (Agent Skill)
- Install (CLI): `npx skillmds@latest add mokhtarabadi/task-generator`
- Raw SKILL.md: https://api.skillmd.com/api/skills/mokhtarabadi/task-generator/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: mokhtarabadi (https://skillmd.com/u/mokhtarabadi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/mokhtarabadi/task-generator

---


# Task Generator Workflow

You are the Task Generator. Your job is to create structured task files for the Manager.

## Workflow

1. **Analyze:** Determine if the request is a `bug`, `improvement`, or `feature`.
2. **Index:** Search all Kanban subdirectories in `tasks/` for the highest existing task ID. Run:

```bash
NEXT_ID=$(find tasks/ -type f -name "*.md" -exec basename {} \; | grep -Eo '^[0-9]+' | sort -n | tail -1 | awk '{print $1+1}')
if [ -z "$NEXT_ID" ] || [ "$NEXT_ID" -eq 0 ] 2>/dev/null; then NEXT_ID="01"; fi
printf "%02d\n" $NEXT_ID
```

Use the output as the zero-padded task number. If `tasks/` doesn't exist, create it along with the Kanban subdirectories and start at `01`.

Title consistency check (unified template `# Task [NN]: [Title]`): flag any duplicate title numbers:

```bash
grep -rhn "^# Task [0-9][0-9]*:" tasks/ | sort | uniq -d
```

The title number MUST match the filename ID. Any mismatch or duplicate must be resolved with the Collision Check below before writing the file.

Duplicate ID check — flag any duplicated numeric task IDs across the ACTIVE Kanban directories only:

```bash
find tasks/backlog tasks/in-progress tasks/qa tasks/completed -type f -name "*.md" -exec basename {} \; | grep -Eo '^[0-9]+' | sort | uniq -d
```

If the output is non-empty, HALT and report duplicate task IDs. Do NOT overwrite files. Archive is a historical record and MUST NOT be included in the blocking duplicate-ID check. If archive duplicates are discovered separately, report them as a warning only, never HALT task creation.

3. **Name:** Create a kebab-case filename (e.g., `01-fix-login-bug.md`). Place it in `tasks/backlog/`.

3.5. **Collision Check:** Before writing the file, verify that `tasks/backlog/{NEXT_ID}-*.md` does NOT already exist. Run: `ls tasks/backlog/${NEXT_ID}-*.md 2>/dev/null`. If a file with that ID already exists, HALT and report: '⚠️ Task ID collision: {NEXT_ID} is already in use. Re-run ID discovery.' Do NOT overwrite existing files.

4. **Generate File:** Write the following unified canonical template to the new file. The `## Source Context` section is **polymorphic** — include ONLY the variant block matching the `**Source:**` value. `## Goal`, `## Local TODOs`, `## Acceptance Criteria`, `## Verification Evidence`, and `## Risk & Rollback` are MANDATORY and UNCONDITIONAL for ALL source types (lint contract):

   ```markdown
   # Task [NN]: [Title]

   **File:** `tasks/backlog/[filename]`
   **Source:** [orchestrator|telegram|manager]
   **Type:** [bug|improvement|feature]
   **Status:** open

   ## Source Context

   ### Variant A: Orchestrator (`**Source:** orchestrator`)

   ## Goal

   [Summary of the goal — MANDATORY]

   ## Blueprint Reference

   [Link or reference to the approved architectural blueprint]

   ## Manager's Notes

   [Any specific notes, requirements, or constraints]

   ### Variant B: Telegram (`**Source:** telegram`)

   ## Goal

   [MANDATORY — one-line goal derived from the RAW_TEXT]

   ## Original Message ([Language])

   {RAW_TEXT — verbatim, zero changes}

   ## English Translation

   {EN_TRANSLATION}

   ## Refactored Prompt

   {REFACTORED_PROMPT}

   ## Relevant Code Context

   {CODEBASE_CONTEXT}

   ## AI Analysis & Opinion

   {AI_OPINION}

   ### Variant C: Manager (`**Source:** manager`)

   ## Goal

   [Summary of the goal — MANDATORY]

   ## Manager's Notes

   [Any specific notes, requirements, or constraints]

   <!-- These sections are unconditional per lint contract — DO NOT move back inside variants -->

   ## Local TODOs

   - [ ] Initial codebase exploration
   - [ ] [Specific step 1]
   - [ ] Verify functionality

   ## Acceptance Criteria

   - [ ] [Criterion 1 — what must be true for this task to be considered done]
   - [ ] [Criterion 2]

   ## Verification Evidence

   - **Test command:** [exact command]
   - **Expected result:** [what success looks like]
   - **Actual result:** _(The Hands fill this during execution)_
   - **Exit code:** _(The Hands fill this during execution)_

   ## Definition of Done

   The task is NOT done unless ALL of the following are true (unconditional, applies to every source type):

   - [ ] Build/Test/Lint pass with exit code 0
   - [ ] `lint_task_file` passes on the active task file
   - [ ] `CHANGELOG.md` updated via Parse-Then-Append
   - [ ] `verification-before-completion` applied and evidence recorded

   > **Box-checking mandate:** During the implementation `<summary_phase>`, the Hands MUST check every `## Acceptance Criteria` and `## Definition of Done` box that is genuinely satisfied by the recorded `## Verification Evidence` — do NOT defer box-checking to a closure task. See `<hands_protocols>` for the authoritative instruction.

   ## Risk & Rollback

   - **Risk:** [what could go wrong]
   - **Rollback plan:** [how to undo if needed]

   ---

   ## Execution Log & Reasoning

   _(The Hands: Manually log your technical changes, file edits, and architectural reasoning here BEFORE calling the MCP tool)_

   ## Factual Git Diff

   <!-- BEGIN_GIT_DIFF -->

   _(Git diff will be automatically injected here by the MCP tool. Do not edit this block manually)_

   <!-- END_GIT_DIFF -->
   ```

## Multi-Phase Task Template

If the Orchestrator specifies `multi_phase: true`, generate a SINGLE task file with inline phase sections instead of separate files. Use this structure (same unified metadata header, `## Goal` mandatory):

```markdown
# Task [NN]: [Title]

**File:** `tasks/backlog/[filename]`
**Source:** [orchestrator|telegram|manager]
**Type:** [bug|improvement|feature]
**Status:** open

## Goal

[Summary of the goal]

## Acceptance Criteria

- [ ] [Criterion 1 — what must be true for this task to be considered done]
- [ ] [Criterion 2]

## Verification Evidence

- **Test command:** [exact command]
- **Expected result:** [what success looks like]
- **Actual result:** _(The Hands fill this during execution)_
- **Exit code:** _(The Hands fill this during execution)_

## Definition of Done

The task is NOT done unless ALL of the following are true (unconditional, applies to every multi-phase task):

- [ ] Build/Test/Lint pass with exit code 0
- [ ] `lint_task_file` passes on the active task file
- [ ] `CHANGELOG.md` updated via Parse-Then-Append
- [ ] `verification-before-completion` applied and evidence recorded

> **Box-checking mandate:** During the implementation `<summary_phase>`, the Hands MUST check every `## Acceptance Criteria` and `## Definition of Done` box that is genuinely satisfied by the recorded `## Verification Evidence` — do NOT defer box-checking to a closure task. See `<hands_protocols>` for the authoritative instruction.

## Risk & Rollback

- **Risk:** [what could go wrong]
- **Rollback plan:** [how to undo if needed]

## Phase 1: [Name]

### Local TODOs

- [ ] [Phase 1 step 1]
- [ ] [Phase 1 step 2]

## Phase 2: [Name]

### Local TODOs

- [ ] [Phase 2 step 1]
- [ ] [Phase 2 step 2]

## Execution Log & Reasoning

_(The Hands: Manually log your technical changes, file edits, and architectural reasoning here BEFORE calling the MCP tool)_

## Factual Git Diff

<!-- BEGIN_GIT_DIFF -->

_(Git diff will be automatically injected here by the MCP tool. Do not edit this block manually)_

<!-- END_GIT_DIFF -->
```

5. **Halt and Handover:** DO NOT execute the task. Print the exact message: "✅ The task file has been created at `tasks/backlog/[filename]` and is ready to be sent to the Orchestrator." and STOP.

## Bundle Workflow (Meta-Tasks) — Task 110/155 (Pure MCP)

Use this when the Manager has 4–6 small related tasks that should be executed together instead of sequentially. The bundler preserves every requirement verbatim and archives the sources.

### When to Use

- Manager says: "bundle tasks 1, 2, 5, 10, 15, 20" or "create a meta-task from 12 15 20"
- Tasks are small, same stack/domain (e.g., all Android polish), and would be inefficient to run one-by-one
- Goal is one branch, one `Factual Git Diff`, one QA gate (all-or-nothing)

### Canonical Invocation — Pure MCP Tool (Task 155)

Invoke the `bundle_tasks` MCP tool via the Hands:

```json
{
  "tool": "bundle_tasks",
  "arguments": {
    "task_ids": ["12", "15", "20"],
    "title": "android-polish-bundle",
    "dry_run": false,
    "force": false
  }
}
```

Examples (MCP arguments):

- `task_ids: ["12","15","20"], title: "android-polish-bundle"`
- `task_ids: ["12","15","20"], title: "android-polish-bundle", dry_run: true`
- `task_ids: ["1","2","3","4","5","6","7"], title: "mega-bundle", force: true  // bypass 6-cap`

### What the Script Does (Deterministic, No LLM)

1. **Validate IDs:** searches `tasks/backlog/ tasks/in-progress/ tasks/qa/ tasks/completed/` (active only, archive excluded) for each `<id>-*.md`. HALT if any missing or if archive already contains the ID (already superseded). Rejects non-numeric IDs.
2. **Next-ID Discovery:** `find tasks -type f -name "*.md" | grep -Eo '^[0-9]+' | sort -n | tail -1 +1` across ALL dirs (including archive) — guarantees no collision. Zero-padded (`02d` for <100, raw for >=100).
3. **Slug:** kebab-case the `--title` (`android Polish_Bundle` → `android-polish-bundle`). Output file: `tasks/backlog/<NEXT_ID>-<slug>.md`.
4. **Verbatim Extraction:** for each source, extracts `## Goal`, `## Manager's Notes`/`## Blueprint Reference`, `## Acceptance Criteria`, `## Local TODOs`, `## Risk & Rollback` verbatim (regex until next `## `). No summarization.
5. **Generate META File:** canonical template + extra metadata:
   - `**Supersedes:** [12, 15, 20]`
   - `**Meta:** true` (flag for tooling; `**Type:**` stays `feature` for lint compat, `meta` is also allowed)
   - Per-source appendix: `### Source Task XX: Title` with verbatim blocks
   - `## Bundled Checklist (All-or-Nothing)` — every source AC line prefixed `[XX]`, single QA gate
   - `## Local TODOs` aggregates all source TODOs prefixed `[XX]` plus bundle-specific steps
   - Guardrail notes: diff-size warning if combined LOC >400
6. **Auto-Archive:** unless `--dry-run`, runs `git mv <source> tasks/archive/<source>` (fallback to `mv` + `git add` for untracked) and patches the archived file:
   - `**File:**` → `tasks/archive/<file>`
   - `**Status:** superseded`
   - `**Superseded-By:** <META_ID>-<slug>` + `**Superseded-At:** YYYY-MM-DD`
   - Superseded footer before `## Execution Log` with `git log --follow` hint
     History remains reachable: `git log --oneline --follow -- tasks/archive/<file>`

### Guardrails

- **Cap:** `MAX_BUNDLE_SIZE=6` — rejects >6 without `--force` (mega-diff prevention)
- **Diff-size:** warns if combined source LOC >400 ("consider split")
- **Duplicate/Collision:** `ls tasks/backlog/<NEXT_ID>-*.md` check before write; HALT if exists. Duplicate active IDs HALT.
- **Archive-Only:** `git mv` to `tasks/archive/` — never `git rm` / purge until META reaches `tasks/completed/`; rollback is `git mv tasks/archive/<id>-*.md tasks/backlog/`

### QA & Completion

- META follows normal Kanban: `tasks/backlog/<META>` → `tasks/in-progress/<META>` → `tasks/qa/<META>` → `tasks/completed/<META>` with a single injected diff
- QA is all-or-nothing: if ANY bundled criterion fails, entire META is `QA_REJECTED`

### Verification

```bash
bundle_tasks(task_ids=["12","15","20"], title="test", dry_run=true)
lint_task_file tasks/backlog/<META_FILE>
git log --oneline --follow -- tasks/archive/12-*.md | head
```