Rewrite onboarding documentation for a defined reader
Reorganize supplied onboarding material around prerequisites, steps, and verifiable completion.
Clarifies that file creation tasks are complete when the deliverable is successfully written—no submission step required
---
name: deliverable-completion
description: Clarifies that file creation tasks are complete when the deliverable is successfully written—no submission step required
---
# Deliverable Completion Protocol
## Purpose
This skill addresses a common misconception during document and file creation tasks: agents sometimes search for or attempt to use a `submit_work`, `finalize`, or similar tool after creating the deliverable. **No such step is required.** Task completion is achieved when the file is successfully created with the required content.
## Core Principle
**File creation = Task completion**
When a task requests you to create a document, report, script, or any file-based deliverable, the task is complete once:
1. The file has been written to disk
2. The file contains the required content
3. The file is in the correct format and location
There is **no additional submission, finalization, or confirmation step** needed.
## Execution Workflow
### Step 1: Create the Deliverable
Use the appropriate file creation method for your task:
```python
# For programmatic file creation
with open('deliverable.docx', 'wb') as f:
f.write(document_content)
```
```bash
# For shell-based creation
echo "content" > output.txt
```
Or use available tools like `write_file`, `create_file`, etc.
### Step 2: Verify Creation
Confirm the file exists and contains expected content:
```bash
ls -la deliverable.docx
# or
cat output.txt
```
### Step 3: Declare Completion
Once verification passes, **the task is complete**. Do not:
- Search for a `submit_work` tool
- Look for a `finalize_task` function
- Attempt to "upload" or "submit" the file elsewhere
- Add extra confirmation steps
Simply report that the deliverable has been created successfully.
## Common Mistakes to Avoid
| ❌ Incorrect | ✅ Correct |
|-------------|-----------|
| Creating file, then searching for submit tool | Creating file, verifying, declaring done |
| Assuming a finalization API exists | Treating file creation as the final step |
| Adding unnecessary confirmation steps | Completing after successful write |
## Example Task Flow
**Task:** "Create a negotiation strategy document covering BATNA, ZOPA, and timeline."
**Correct Execution:**
1. Write the document content
2. Save as `negotiation_strategy.docx`
3. Verify file exists (~43KB, contains all sections)
4. Report: "Negotiation strategy document created successfully"
5. **Task complete** — no further action needed
## When This Applies
- Document creation (.docx, .pdf, .md, .txt)
- Code file generation (.py, .sh, .js)
- Data exports (.csv, .json, .xlsx)
- Configuration files (.yaml, .toml, .ini)
- Any file-based deliverable
## When This Does NOT Apply
- Tasks explicitly requiring external submission (e.g., "submit to API endpoint")
- Tasks requiring human review/approval workflows
- Tasks where the file is an intermediate step (not the final deliverable)---
name: deliverable-completion
description: Clarifies that file creation tasks are complete when the deliverable is successfully written—no submission step required
---
# Deliverable Completion Protocol
## Purpose
This skill addresses a common misconception during document and file creation tasks: agents sometimes search for or attempt to use a `submit_work`, `finalize`, or similar tool after creating the deliverable. **No such step is required.** Task completion is achieved when the file is successfully created with the required content.
## Core Principle
**File creation = Task completion**
When a task requests you to create a document, report, script, or any file-based deliverable, the task is complete once:
1. The file has been written to disk
2. The file contains the required content
3. The file is in the correct format and location
There is **no additional submission, finalization, or confirmation step** needed.
## Execution Workflow
### Step 1: Create the Deliverable
Use the appropriate file creation method for your task:
```python
# For programmatic file creation
with open('deliverable.docx', 'wb') as f:
f.write(document_content)
```
```bash
# For shell-based creation
echo "content" > output.txt
```
Or use available tools like `write_file`, `create_file`, etc.
### Step 2: Verify Creation
Confirm the file exists and contains expected content:
```bash
ls -la deliverable.docx
# or
cat output.txt
```
### Step 3: Declare Completion
Once verification passes, **the task is complete**. Do not:
- Search for a `submit_work` tool
- Look for a `finalize_task` function
- Attempt to "upload" or "submit" the file elsewhere
- Add extra confirmation steps
Simply report that the deliverable has been created successfully.
## Common Mistakes to Avoid
| ❌ Incorrect | ✅ Correct |
|-------------|-----------|
| Creating file, then searching for submit tool | Creating file, verifying, declaring done |
| Assuming a finalization API exists | Treating file creation as the final step |
| Adding unnecessary confirmation steps | Completing after successful write |
## Example Task Flow
**Task:** "Create a negotiation strategy document covering BATNA, ZOPA, and timeline."
**Correct Execution:**
1. Write the document content
2. Save as `negotiation_strategy.docx`
3. Verify file exists (~43KB, contains all sections)
4. Report: "Negotiation strategy document created successfully"
5. **Task complete** — no further action needed
## When This Applies
- Document creation (.docx, .pdf, .md, .txt)
- Code file generation (.py, .sh, .js)
- Data exports (.csv, .json, .xlsx)
- Configuration files (.yaml, .toml, .ini)
- Any file-based deliverable
## When This Does NOT Apply
- Tasks explicitly requiring external submission (e.g., "submit to API endpoint")
- Tasks requiring human review/approval workflows
- Tasks where the file is an intermediate step (not the final deliverable)Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.