AutocontextSAFE

a recursive self-improving harness designed to help your agents (and future iterations of those agents) succeed on any task

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: autocontext
description: >
  Iterative strategy generation and evaluation system. Use when the user wants
  to evaluate agent output quality, run improvement loops, queue tasks for
  background evaluation, check run status, inspect runtime artifacts and
  session branch lineage, or discover available scenarios. Provides LLM-based
  judging with rubric-driven scoring.
allowed-tools: autocontext_judge autocontext_improve autocontext_status autocontext_scenarios autocontext_queue autocontext_runtime_snapshot
---

# autocontext

autocontext is an iterative strategy generation and evaluation system that uses
LLM-based judging to score and improve agent outputs.

## Available Tools

- **autocontext_judge** — Evaluate agent output against a rubric. Returns a 0–1
  score with reasoning and per-dimension breakdowns.
- **autocontext_improve** — Run a multi-round improvement loop. The agent output
  is judged, revised based on feedback, and re-evaluated until the quality
  threshold is met or max rounds are exhausted.
- **autocontext_queue** — Enqueue a task for background evaluation by the task
  runner daemon.
- **autocontext_status** — Check the status of runs and queued tasks.
- **autocontext_scenarios** — List available evaluation scenarios and their
  families.
- **autocontext_runtime_snapshot** — Inspect run artifacts, package provenance,
  branchable session lineage, and recent event-stream entries.

## Quick Start

### 1. Evaluate output quality

Use `autocontext_judge` with a task prompt, the agent's output, and a rubric:

```
autocontext_judge(
  task_prompt="Write a Python function to parse CSV files",
  agent_output="def parse_csv(path): ...",
  rubric="Correctness, error handling, edge cases, documentation"
)
```

### 2. Improve output iteratively

Use `autocontext_improve` to automatically revise output through
judge-guided feedback loops:

```
autocontext_improve(
  task_prompt="Write a Python function to parse CSV files",
  initial_output="def parse_csv(path): ...",
  rubric="Correctness, error handling, edge cases, documentation",
  max_rounds=5,
  quality_threshold=0.85
)
```

### 3. Queue background tasks

Use `autocontext_queue` with a scenario name to enqueue evaluation tasks
for asynchronous processing:

```
autocontext_queue(spec_name="my_scenario")
```

Check results later with `autocontext_status`.

For deeper context, use `autocontext_runtime_snapshot` with the run ID. Add
`session_id` when you need the active branch path before continuing work:

```
autocontext_runtime_snapshot(run_id="run_123", session_id="sess_123")
```

### 4. Discover scenarios

Use `autocontext_scenarios` to see what evaluation scenarios are available:

```
autocontext_scenarios()
autocontext_scenarios(family="agent_task")
```

## Configuration

The extension auto-detects configuration from these sources:

1. **Project config** — `.autoctx.json` in the working directory (created via `autoctx init`)
2. **Environment variables:**
   - `AUTOCONTEXT_AGENT_PROVIDER` or `AUTOCONTEXT_PROVIDER` — Provider type
   - `AUTOCONTEXT_AGENT_API_KEY` or `AUTOCONTEXT_API_KEY` — Provider API key
   - `AUTOCONTEXT_AGENT_DEFAULT_MODEL` or `AUTOCONTEXT_MODEL` — Model override
   - `AUTOCONTEXT_DB_PATH` — SQLite database path override
3. **Pi provider** — Falls back to Pi's configured LLM provider

## CLI Companion

For standalone usage outside Pi, install the `autoctx` CLI:

```bash
npm install -g autoctx
autoctx init
autoctx solve "your problem" --iterations 5
autoctx simulate --description "your simulation" --runs 3
```

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.