Skip to content
Back to skills

Adhoc

ASecurity

Run and manage ad-hoc AI prompts in sgai workspaces without starting a full agentic session. Use when you need to run a one-off AI prompt against a workspace, check the status of a running ad-hoc prompt, or stop a running ad-hoc prompt.

  • 57 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
ai-agentsgobashapisecuritydocumentation

Works with

  • api

Security analysis

A100/100

Scanned October 17, 2026

npx -y skills add diegosouzapw/awesome-omni-skill --skill adhoc --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Adhoc?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Adhoc
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/diegosouzapw-adhoc/badge)](https://www.skillsdirectory.com/skills/diegosouzapw-adhoc)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: adhoc
description: Run and manage ad-hoc AI prompts in sgai workspaces without starting a full agentic session. Use when you need to run a one-off AI prompt against a workspace, check the status of a running ad-hoc prompt, or stop a running ad-hoc prompt.
compatibility: Requires a running sgai server and opencode installed. The prompt runs as a non-interactive opencode command in the workspace directory.
---

# Ad-hoc Prompts

Ad-hoc prompts let you run a single AI prompt against a workspace without starting a full agentic session. Useful for quick tasks, code reviews, or one-off questions.

## Start an Ad-hoc Prompt

**Endpoint:** `POST /api/v1/workspaces/{name}/adhoc`

```bash
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Review the authentication code and identify any security issues",
    "model": "anthropic/claude-opus-4-6"
  }'
```

Request:
```json
{
  "prompt": "Review the authentication code and identify any security issues",
  "model": "anthropic/claude-opus-4-6"
}
```

Response:
```json
{
  "running": true,
  "output": "",
  "message": "ad-hoc prompt started"
}
```

If already running:
```json
{
  "running": true,
  "output": "$ opencode run -m anthropic/claude-opus-4-6...\nprompt: Review the...",
  "message": "ad-hoc prompt already running"
}
```

### Model Format

The model parameter accepts the same format as GOAL.md models:
- `"anthropic/claude-opus-4-6"` — standard model
- `"anthropic/claude-opus-4-6 (max)"` — model with variant
- Use `GET /api/v1/models` to list available models

### How Ad-hoc Works

Ad-hoc prompts run `opencode run -m {model} --agent build --title "adhoc [{model}]"` with your prompt piped as stdin. The workspace directory is used as the working directory.

## Get Ad-hoc Status

Check if an ad-hoc prompt is running and get its current output.

**Endpoint:** `GET /api/v1/workspaces/{name}/adhoc`

```bash
curl -s $BASE_URL/api/v1/workspaces/my-project/adhoc
```

Response (running):
```json
{
  "running": true,
  "output": "$ opencode run -m anthropic/claude-opus-4-6 --agent build --title \"adhoc [anthropic/claude-opus-4-6]\"\nprompt: Review the authentication code...\n\nAnalyzing the codebase...\n\nFound 3 potential issues:\n1. JWT tokens lack expiration...",
  "message": "adhoc status"
}
```

Response (not running):
```json
{
  "running": false,
  "output": "$ opencode run...\n...\n[completed successfully]",
  "message": "adhoc status"
}
```

### Polling for Completion

```bash
WORKSPACE="my-project"

# Start the prompt
curl -X POST $BASE_URL/api/v1/workspaces/$WORKSPACE/adhoc \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Summarize the GOAL.md", "model": "anthropic/claude-sonnet-4-6"}'

# Poll until done
while true; do
  STATUS=$(curl -s $BASE_URL/api/v1/workspaces/$WORKSPACE/adhoc)
  RUNNING=$(echo $STATUS | jq '.running')
  
  if [ "$RUNNING" = "false" ]; then
    echo "Ad-hoc complete!"
    echo $STATUS | jq -r '.output'
    break
  fi
  
  echo "Still running..."
  sleep 3
done
```

## Stop an Ad-hoc Prompt

Stop a running ad-hoc prompt.

**Endpoint:** `DELETE /api/v1/workspaces/{name}/adhoc`

```bash
curl -X DELETE $BASE_URL/api/v1/workspaces/my-project/adhoc
```

Response:
```json
{
  "running": false,
  "output": "$ opencode run...\n...\n[process terminated]",
  "message": "ad-hoc stopped"
}
```

## Ad-hoc vs Full Session

| Feature | Ad-hoc | Full Session |
|---------|--------|--------------|
| Single prompt | ✓ | ✗ (multi-agent flow) |
| Multi-agent workflow | ✗ | ✓ |
| Human interaction | ✗ | ✓ |
| Progress tracking | Basic | Full (todos, events) |
| Cost tracking | ✗ | ✓ |
| Concurrent with session | No | N/A |
| GOAL.md required | ✗ | ✓ |

## Common Ad-hoc Use Cases

```bash
# Code review
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "Review all Go files for potential race conditions", "model": "anthropic/claude-opus-4-6"}'

# Documentation generation
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "Generate a README.md for this project", "model": "anthropic/claude-sonnet-4-6"}'

# Quick fix
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "Fix the failing test in auth_test.go", "model": "anthropic/claude-sonnet-4-6"}'

# Analysis
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "List all TODO comments in the codebase", "model": "anthropic/claude-haiku-4-5"}'
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.