How do I install the Autocontext skill?

Run npx skillmds@latest add zereight/autocontext in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Autocontext safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Autocontext?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Autocontext free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Autocontext?

zereight (@zereight) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Autocontext by zereight — Agent Skill

Name: Autocontext
Author: zereight

Autocontext

Run iterative agent evaluation and improvement loops (judge, improve, scenarios, playbooks) via pi-autocontext or autoctx CLI. Use when improving repeatable agent workflows (review rubrics, skill quality, scenario-based feedback), checking run status, or accumulating knowledge—not for one-off code edits or repo pattern drift (use continuity) or single PR review (use zereight-review).

zereight 6989c58 3.2 KB Updated Sep 17, 2026

File contents

autocontext (skill repo)

Orchestration layer for evaluate → improve → persist agent outputs. Installed stack:

Layer	Install	Role
Pi	`npm:pi-autocontext` in `~/.pi/agent/settings.json`	In-session tools + `/skill:autocontext`
CLI	`uv tool install autocontext==0.5.0` → `autoctx`	Background `solve` / `run` / `list`
Project	`.autoctx.json` in repo root	Defaults: `agent_task`, `provider: pi`, `gens: 2`

When to use which skill

Need	Use
PR diff review, OWASP, RN effects	`zereight-review`
Codebase pattern SSOT / drift	`continuity`
Chat mistake → durable rule	`session-lessons`
Rubric-driven loop, playbooks, scenario bench	autocontext
Pi-native solve loop + knowledge persistence	harness (faster, no Python dep)

Pi tools (after `pi install npm:pi-autocontext`)

autocontext_judge — score output vs rubric (0–1 + reasoning)
autocontext_improve — judge-guided revision rounds
autocontext_status — runs / queue
autocontext_scenarios — list scenarios
autocontext_queue — background evaluation
autocontext_runtime_snapshot — artifacts, session lineage

Invoke packaged skill: /skill:autocontext (from pi-autocontext package).

CLI (from repo root)

cd /path/to/skill   # this repo
export AUTOCONTEXT_AGENT_PROVIDER=pi
export AUTOCONTEXT_PI_COMMAND=pi

autoctx list
# TypeScript CLI (Pi npm): ~/.pi/agent/npm/node_modules/.bin/autoctx capabilities

# Full loop (costs Pi/Cursor quota — start small)
autoctx solve "Improve zereight-review checklist consistency for RN PRs" --iterations 2

Artifacts: runs/, knowledge/ (gitignored). DB default: runs/autocontext.sqlite3.

First scenario for this repo

Goal: Make zereight-review feedback more consistent on sample PR diffs.

Rubric: actionable findings, Effect anti-patterns, security mention when relevant, no vague praise-only comments.
Judge: autocontext_judge with task = review prompt + diff snippet, output = draft review.
Improve: Hints that pass judge → zereight-review/references/ or checklist (human approves merge).
Stop: gens: 2 in .autoctx.json; raise only after reviewing runs/*/report.md.

Cost and safety

provider: pi uses your Pi default (cursor/composer-2.5) — each generation consumes quota.
Start with --iterations 1 or autocontext_judge only before full solve.
Do not install npm package autocontext (wrong project). Use pi-autocontext / PyPI autocontext / npm autoctx.

Env reference

Copy autocontext.env.example and export in shell or Pi session. Key vars:

AUTOCONTEXT_AGENT_PROVIDER=pi
AUTOCONTEXT_PI_COMMAND=pi
AUTOCONTEXT_DB_PATH=runs/autocontext.sqlite3
AUTOCONTEXT_DEFAULT_GENERATIONS=2

---name: autocontextdescription: Run iterative agent evaluation and improvement loops (judge, improve, scenarios, playbooks) via pi-autocontext or autoctx CLI. Use when improving repeatable agent workflows (review rubrics, skill quality, scenario-based feedback), checking run status, or accumulating knowledge—not for one-off code edits or repo pattern drift (use continuity) or single PR review (use zereight-review).---# autocontext (skill repo)Orchestration layer for **evaluate → improve → persist** agent outputs. Installed stack:| Layer | Install | Role ||-------|---------|------|| Pi | `npm:pi-autocontext` in `~/.pi/agent/settings.json` | In-session tools + `/skill:autocontext` || CLI | `uv tool install autocontext==0.5.0` → `autoctx` | Background `solve` / `run` / `list` || Project | `.autoctx.json` in repo root | Defaults: `agent_task`, `provider: pi`, `gens: 2` |## When to use which skill| Need | Use ||------|-----|| PR diff review, OWASP, RN effects | `zereight-review` || Codebase pattern SSOT / drift | `continuity` || Chat mistake → durable rule | `session-lessons` || Rubric-driven loop, playbooks, scenario bench | **autocontext** || Pi-native solve loop + knowledge persistence | **harness** (faster, no Python dep) |## Pi tools (after `pi install npm:pi-autocontext`)- `autocontext_judge` — score output vs rubric (0–1 + reasoning)- `autocontext_improve` — judge-guided revision rounds- `autocontext_status` — runs / queue- `autocontext_scenarios` — list scenarios- `autocontext_queue` — background evaluation- `autocontext_runtime_snapshot` — artifacts, session lineageInvoke packaged skill: `/skill:autocontext` (from `pi-autocontext` package).## CLI (from repo root)```bashcd /path/to/skill # this repoexport AUTOCONTEXT_AGENT_PROVIDER=piexport AUTOCONTEXT_PI_COMMAND=piautoctx list# TypeScript CLI (Pi npm): ~/.pi/agent/npm/node_modules/.bin/autoctx capabilities# Full loop (costs Pi/Cursor quota — start small)autoctx solve "Improve zereight-review checklist consistency for RN PRs" --iterations 2```Artifacts: `runs/`, `knowledge/` (gitignored). DB default: `runs/autocontext.sqlite3`.## First scenario for this repo**Goal:** Make `zereight-review` feedback more consistent on sample PR diffs.1. **Rubric:** actionable findings, Effect anti-patterns, security mention when relevant, no vague praise-only comments.2. **Judge:** `autocontext_judge` with task = review prompt + diff snippet, output = draft review.3. **Improve:** Hints that pass judge → `zereight-review/references/` or checklist (human approves merge).4. **Stop:** `gens: 2` in `.autoctx.json`; raise only after reviewing `runs/*/report.md`.## Cost and safety- `provider: pi` uses your Pi default (`cursor/composer-2.5`) — **each generation consumes quota**.- Start with `--iterations 1` or `autocontext_judge` only before full `solve`.- Do not install npm package `autocontext` (wrong project). Use `pi-autocontext` / PyPI `autocontext` / npm `autoctx`.## Env referenceCopy `autocontext.env.example` and export in shell or Pi session. Key vars:- `AUTOCONTEXT_AGENT_PROVIDER=pi`- `AUTOCONTEXT_PI_COMMAND=pi`- `AUTOCONTEXT_DB_PATH=runs/autocontext.sqlite3`- `AUTOCONTEXT_DEFAULT_GENERATIONS=2`

Frequently asked questions