Skip to content

Commit 2a36944

Browse files
author
Corellis
committed
Add P1 #3: task-autopilot skill — automatic task decomposition and planning
1 parent bfaa5ea commit 2a36944

1 file changed

Lines changed: 174 additions & 0 deletions

File tree

‎templates/skills/task-autopilot/SKILL.md‎

Lines changed: 174 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,174 @@
1+
---
2+
name: task-autopilot
3+
slug: task-autopilot
4+
description: "Automatic task decomposition and execution planning. Receives a high-level task, breaks it into subtasks, classifies each (code/research/design/ops), estimates effort, identifies dependencies, and generates a structured plan for owner approval. Works standalone or as part of goal-ops workflows."
5+
---
6+
7+
# Task Autopilot — Automatic Task Decomposition
8+
9+
> Big task in → Structured plan out → Execute on approval
10+
>
11+
> Takes a vague or complex task and produces a concrete, actionable plan
12+
> with subtasks, classifications, effort estimates, and dependency ordering.
13+
14+
## When to Trigger
15+
16+
- **Goal-ops integration**: Controller assigns you a sub-goal (SG) — decompose it into subtasks
17+
- **Direct assignment**: Owner says "do X" where X is complex (multiple steps, multiple concerns)
18+
- **Proactive engine**: After proactive-task-engine identifies a task, autopilot decomposes it
19+
- **Manual**: "break this down" / "plan this out" / "decompose this task"
20+
21+
**Skip autopilot when**: Task is simple and single-step (just do it directly)
22+
23+
## Decomposition Flow
24+
25+
```
26+
┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐
27+
│ Parse │ ──▶ │ Decompose │ ──▶ │ Classify │ ──▶ │ Present │
28+
│ Input │ │ into Steps │ │ & Score │ │ Plan │
29+
└────────────┘ └────────────┘ └────────────┘ └────────────┘
30+
```
31+
32+
## Phase 1: Parse Input
33+
34+
Extract from the task description:
35+
36+
| Field | Source | Default |
37+
|-------|--------|---------|
38+
| Objective | Task description | Required |
39+
| Deadline | Explicit date or "ASAP" / "this week" | None |
40+
| Constraints | Budget, tech stack, dependencies | None |
41+
| Acceptance criteria | Explicit deliverables | Infer from objective |
42+
| Context | Related threads, docs, prior work | Search memory |
43+
44+
If acceptance criteria are missing, generate them and ask for confirmation.
45+
46+
## Phase 2: Decompose
47+
48+
Break the task into subtasks. Rules:
49+
50+
1. **Each subtask should be completable in one work session** (< 4 hours)
51+
2. **Each subtask has a single clear deliverable** (a file, a PR, a report, a config)
52+
3. **Minimize dependencies between subtasks** — prefer parallel execution
53+
4. **Include verification steps** — don't just "write code", also "test code"
54+
55+
### Decomposition Strategy
56+
57+
| Task Size | Approach |
58+
|-----------|----------|
59+
| Small (< 2h) | 2-3 subtasks max, often just do/verify |
60+
| Medium (2-8h) | 4-6 subtasks with clear phases |
61+
| Large (> 8h) | Flag to owner — may need goal-ops level orchestration |
62+
63+
## Phase 3: Classify & Score
64+
65+
For each subtask, assign:
66+
67+
### Classification
68+
69+
| Type | Description | Typical Tools |
70+
|------|-------------|---------------|
71+
| `code` | Write/modify code, create PRs | ACP agents, exec, git |
72+
| `research` | Investigate options, read docs, analyze data | web_search, web_fetch, read |
73+
| `design` | UI/UX specs, architecture decisions, diagrams | browser, canvas, design tools |
74+
| `ops` | Deploy, configure, monitor, debug | exec, docker, ssh |
75+
| `comms` | Write docs, send updates, coordinate | message, write |
76+
77+
### Effort Estimate
78+
79+
| Label | Time | Description |
80+
|-------|------|-------------|
81+
| `trivial` | < 15 min | Config change, simple lookup |
82+
| `small` | 15-60 min | Single-file change, short research |
83+
| `medium` | 1-3 hours | Multi-file feature, detailed research |
84+
| `large` | 3-8 hours | Complex feature, integration work |
85+
86+
### Confidence Score (1-10)
87+
88+
How confident are you that you can complete this subtask successfully?
89+
90+
- **8-10**: Done this before, clear path
91+
- **5-7**: Mostly clear, some unknowns
92+
- **1-4**: Significant unknowns, may need help
93+
94+
## Phase 4: Present Plan
95+
96+
Output a structured plan for approval:
97+
98+
```markdown
99+
📋 **Task Plan: [Task Title]**
100+
101+
**Objective**: [1-sentence summary]
102+
**Total Effort**: ~X hours
103+
**Deadline**: [date or "none"]
104+
105+
| # | Subtask | Type | Effort | Confidence | Depends On |
106+
|---|---------|------|--------|------------|------------|
107+
| 1 | Research existing solutions | research | small | 9/10 | — |
108+
| 2 | Design API schema | design | medium | 8/10 | #1 |
109+
| 3 | Implement backend endpoints | code | medium | 7/10 | #2 |
110+
| 4 | Write integration tests | code | small | 8/10 | #3 |
111+
| 5 | Deploy to staging | ops | trivial | 9/10 | #4 |
112+
| 6 | Update documentation | comms | small | 9/10 | #3 |
113+
114+
**Execution Order**:
115+
- Parallel: #1 can start immediately
116+
- Sequential: #2 → #3 → #4 → #5
117+
- Parallel: #6 can start after #3
118+
119+
**Risks**:
120+
- #3 depends on [external API] — if unavailable, will mock and flag
121+
122+
**Acceptance Criteria**:
123+
- [ ] All endpoints return correct responses
124+
- [ ] Test coverage > 80%
125+
- [ ] Deployed and accessible on staging
126+
127+
Approve this plan? (reply "go" / "adjust #3 to ..." / "skip #6")
128+
```
129+
130+
## After Approval
131+
132+
1. **Update task board** (if integrated): Create subtask entries
133+
2. **Execute sequentially**: Complete each subtask, report progress
134+
3. **Checkpoint after each subtask**: Brief status update in thread
135+
4. **On completion**: Summary report with all deliverables
136+
137+
## Integration with Goal-Ops
138+
139+
When used within a goal-ops workflow:
140+
141+
1. Controller assigns SG → triggers task-autopilot
142+
2. Autopilot decomposes → posts plan in SG thread
143+
3. Controller (or owner) approves → lobster executes
144+
4. Subtask completions update the task board
145+
5. All subtasks done → lobster reports to controller for acceptance
146+
147+
## Configuration
148+
149+
Optional `autopilot-config.json` in workspace:
150+
151+
```json
152+
{
153+
"autoDecompose": true,
154+
"maxSubtasks": 8,
155+
"requireApproval": true,
156+
"defaultTaskBoard": "notion",
157+
"confidenceThreshold": 5
158+
}
159+
```
160+
161+
## File Structure
162+
163+
```
164+
skills/task-autopilot/
165+
├── SKILL.md # This file
166+
└── references/
167+
└── classification-guide.md # Detailed type/effort classification examples
168+
```
169+
170+
## Dependencies
171+
172+
- Task board integration (optional, for creating subtask entries)
173+
- goal-participant skill (when used within goal-ops)
174+
- Owner/controller for plan approval

0 commit comments

Comments
 (0)