forge

The operational layer for coding agents. Memory, validation, and feedback loops that compound between sessions.

Installation
CLI
npx skills add https://github.com/boshu2/agentops --skill forge

Installieren Sie diesen Skill über die CLI und beginnen Sie mit der Verwendung des SKILL.md-Workflows in Ihrem Arbeitsbereich.

Zuletzt aktualisiert am 5/5/2026

AgentOps

Validate
Nightly
GitHub stars

Operational discipline for coding agents.

Ship reliable code with unreliable agents.

AgentOps is the context compiler — compile context, gate output, compound knowledge across Claude, Codex, Cursor, and OpenCode.

Install · See It Work · Quick Start · Why DevOps? · Skills · CLI · Doctrine · Docs


Install

Pick the runtime you use.

Claude Code

claude plugin marketplace add boshu2/agentops
claude plugin install agentops@agentops-marketplace

Codex CLI on macOS, Linux, or WSL

curl -fsSL https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install-codex.sh | bash

Codex CLI on Windows PowerShell

irm https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install-codex.ps1 | iex

OpenCode

curl -fsSL https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install-opencode.sh | bash

Other skills-compatible agents

npx skills@latest add boshu2/agentops --cursor -g

Restart your agent after install. Then type /quickstart in your agent chat.

The ao CLI is optional, but recommended. It unlocks repo-native bookkeeping, retrieval, health checks, and terminal workflows.

macOS

brew tap boshu2/agentops https://github.com/boshu2/homebrew-agentops
brew install agentops
ao version

Windows PowerShell

irm https://raw.githubusercontent.com/boshu2/agentops/main/scripts/install-ao.ps1 | iex
ao version

You can also install the CLI from release binaries or build from source.

Concern Answer
What it touches Installs skills globally and registers runtime hooks when requested; agent work writes local bookkeeping to .agents/
Source code changes None during install
Network behavior Install and update paths fetch from GitHub; repo artifacts stay local unless you choose external tools or remote model runtimes
Telemetry None required
Permission surface Skills can run shell commands and read or write repo files during agent work, so install where you want agents to operate
Reversible Remove the installed skill directories, delete .agents/, and remove hook entries from your runtime settings

Troubleshooting: docs/troubleshooting.md · Configuration: docs/ENV-VARS.md


See It Work

One command: validate a PR across Claude and Codex

> /council --mixed validate this PR

[council] evidence packet sealed -> 6 judges across Claude Code and Codex CLI
[claude/judge-1] WARN - rate limiting missing on /login endpoint
[claude/judge-2] PASS - Redis integration follows middleware pattern
[codex/judge-1]  WARN - token bucket refill lacks jitter under burst
[codex/judge-2]  PASS - backoff bounds match retry policy
Consensus: WARN - fix /login rate limit and add refill jitter before shipping
Recorded: .agents/council/<run-id>/verdict.md

Full loop: research through post-mortem

> /rpi "add retry backoff to rate limiter"

[research]    Found 3 prior learnings on rate limiting
[plan]        2 issues, 1 wave
[pre-mortem]  Council validates the plan
[crank]       Executes the scoped work
[vibe]        Council validates the code
[post-mortem] Captures new learnings in .agents/
[flywheel]    Next session starts with better context

The point is not a bigger prompt. The point is a repo that remembers what worked.


What AgentOps Gives You

Three layers. Each solves a different problem. All three compound.

Layer Problem What changes
Context Compiler Every session starts from zero ao inject delivers decay-ranked knowledge. ao context assemble builds phase-scoped packets. 71 skills load automatically via hooks. Your agent starts loaded, not cold.
Validation Gates Agents ship confident garbage /pre-mortem, /vibe, /council — multi-model consensus validates plans before build and code before commit. Gates block, not advise. Three fresh judges catch what one agent can't.
Knowledge Flywheel Lessons disappear between sessions /forge extracts learnings. ao flywheel close-loop scores and promotes. /evolve fixes the worst gap autonomously. /dream compounds overnight. Session 15 starts with everything session 1 learned.
flowchart LR
    S[Session starts] --> L1[Layer 1: Compile context]
    L1 --> W[Agent works]
    W --> L2[Layer 2: Validate output]
    L2 --> E[Session ends]
    E --> L3[Layer 3: Extract + compound]
    L3 -->|better context| S

All state lives in local .agents/ — plain text you can grep, diff, and review. Zero telemetry. Zero cloud dependency. Runtime-neutral across Claude Code, Codex CLI, and OpenCode.

The three gaps (proof contract)

The three layers close three failure modes most agent setups don't even name:

Gap Where the loop breaks Closed by
Judgment Plan looks coherent. Code passes tests. Both miss the edge case. Layer 2: /pre-mortem · /vibe · /council
Durable Learning Auth bug fixed Monday. Same bug returns Wednesday. Layer 3: /forge · ao flywheel · ao inject
Loop Closure Code lands. No lesson extracted. Next session re-learns from scratch. Layer 3→1: /evolve · finding compiler · /dream

Each factor in the 12-factor doctrine closes one or more of these gaps. Full contract: docs/context-lifecycle.md.


Why DevOps?

DevOps proved that disciplined systems around indeterministic workers produce reliable output. SRE proved it again with SLOs and error budgets. Kubernetes proved it for infrastructure with control loops. Coding agents are the next indeterministic worker class. Same playbook. New substrate.

DevOps had the SDLC — the infinity loop that made software delivery an engineering discipline. Coding agents need an equivalent: the Context Development Life Cycle (CDLC). Every SDLC phase has a context counterpart. AgentOps implements all of them.

SDLC CDLC AgentOps surface
Plan Generate /research, /plan, SKILL.md authoring
Code + Build Compile ao context assemble, ao inject, decay-ranked retrieval
Test Test /pre-mortem, /vibe, /council, ao eval run
Release Distribute Skills registry, /converter, cross-runtime export
Deploy Deliver SessionStart hooks, ao inject --for=<skill>
Operate Observe Citation tracking, quality signals, session-outcome
Monitor → Plan Adapt MemRL feedback, /forge, /evolve, /dream

LLMs are engines. Context is fuel. You can't tune the engine — that's the model vendor's job. But you can engineer the fuel. The CDLC is how.

Each generation of software practice gave teams a durable artifact: the wiki, the runbook, the postmortem, the toil budget. AgentOps gives your agents the same kind of artifact — the corpus. A typed, versioned, agent-readable knowledge store maintained alongside the code. The model stays the same. The corpus compounds.

The harness layer commoditizes. Memory primitives, learning loops, validation gates — frontier vendors will ship them natively. What stays yours is the corpus. AgentOps is the bridge tool that helps you build that moat now. See PRODUCT.md for the full thesis.

Full CDLC treatment: docs/cdlc.md. Theoretical foundations: docs/the-science.md and docs/brownian-ratchet.md.


Quick Start

Inside a repo, use the path that matches what you are trying to do.

Path Run Done when
First repo setup ao quick-start, then /quickstart AgentOps reports repo readiness and a next action
First validated change /rpi "a small goal" Discovery, implementation, validation, and learning closeout leave evidence in .agents/
Review something now /council validate this PR or /vibe recent You get a consolidated verdict and an evidence record in .agents/ before shipping

New project? Use the guided CLI seed first:

ao quick-start     # Canonical
ao quickstart      # Stable alias

That command applies the repeatable core seed: .agents/, GOALS.md,
AgentOps instructions, starter knowledge, and readiness guidance. Use
/bootstrap after that when you want the product/operations layer:
PRODUCT.md, README.md, PROGRAM.md/AUTODEV.md, and optional hooks.

Already installed? Ask your agent for the next action:

/quickstart

If you installed the CLI, check your local setup:

ao doctor
ao demo

Full catalog: docs/SKILLS.md · Unsure what to run? Skill Router


Skills

Every skill works alone. Flows compose them when you want more structure.

Skill Use it when
/quickstart You want the fastest setup check and next action
/council You want independent judges — optionally across Claude and Codex — to evaluate one evidence packet and return a consolidated verdict
/research You need codebase context and prior learnings before changing code
/pre-mortem You want to pressure-test a plan before implementation
/implement You want one scoped task built and validated
/rpi You want discovery, build, validation, and bookkeeping in one flow
/vibe You want a code-quality and risk review before shipping
/evolve You want a goal-driven improvement loop with regression gates
/dream You want overnight knowledge compounding that never mutates source code
Full catalog — organized by product layer

Context Compiler (L1): /research · /compile · /inject · /handoff · /recover

Validation Gates (L2): /council · /vibe · /pre-mortem · /post-mortem

Knowledge Flywheel (L3): /retro · /forge · /flywheel · /dream

Orchestrated Flows: /plan · /implement · /crank · /swarm · /rpi · /evolve

Session & Product: /status · /trace · /provenance · /product · /goals · /release · /readme · /doc

Utility: /brainstorm · /bug-hunt · /complexity · /scaffold · /push

Full reference: docs/SKILLS.md

Cross-runtime orchestration — mix Claude, Codex, Cursor, and OpenCode

Multi-runtime, one workflow. The same three layers — context compilation, validation gates, and knowledge flywheel — run whether the active worker is Claude Code, Codex, Cursor, or OpenCode.

One runtime leads a session. Another reviews the result. A third handles focused implementation. Adapters are runtime-specific. The contract is constant: independent context, auditable files, validation before promotion.


The ao CLI

The ao CLI is the repo-native control plane behind the skills. It handles retrieval, health checks, compounding, goals, and terminal workflows.

ao quick-start                            # Set up AgentOps in a repo
ao quickstart                             # Alias for quick-start
ao doctor                                 # Check local health
ao demo                                   # See the value path in 5 minutes
ao search "query"                         # Search session history and local knowledge
ao lookup --query "topic"                 # Retrieve curated learnings and findings
ao context assemble                       # Build a task briefing
ao rpi phased "fix auth startup"          # Run the phased lifecycle from the terminal
ao evolve --max-cycles 1                  # Run one autonomous improvement cycle
ao overnight setup                        # Prepare private Dream runs
ao metrics health                         # Show flywheel health

Full reference: CLI Commands


Advanced: Day Loop And Night Loop

Use /evolve when you want code improvement. It reads GOALS.md, fixes the worst fitness gap, runs regression gates, and records the cycle.

> /evolve

[evolve] GOALS.md loaded
[cycle-1] Worst gap selected
[rpi]     Implements the fix
[gate]    Tests and quality checks pass
[learn]   Post-mortem feeds the flywheel

Use /dream when you want knowledge compounding. It runs offline-style bookkeeping work over .agents/, reports what changed, and never mutates source code, invokes /rpi, or performs git operations.

> /dream start

[overnight] INGEST  harvest new artifacts
[overnight] REDUCE  dedup, defrag, close loops
[overnight] MEASURE corpus quality
[halted]    plateau reached

Morning report: .agents/overnight/<run-id>/summary.md

Run Dream overnight, then run Evolve in the morning against a fresher corpus. The model may be the same; the environment is smarter.


Competitive Positioning

Most tools optimize work within a session. AgentOps compounds across them. The three product layers — Context Compiler, Validation Gates, Knowledge Flywheel — are the gap.

Tool What it does well What AgentOps adds
GSD Fresh-context phased execution, recovery loops, runtime breadth Context Compiler (cross-session retrieval), Validation Gates (pre-build), Knowledge Flywheel
Compound Engineer Ideation, configurable reviewers, cross-runtime conversion Automatic capture/scoring/injection, council validation, repo-native ao workflows
Spec Kit / Kiro Spec-driven development and executable planning artifacts Learning beyond specs: failures, decisions, retros, prevention rules
Superpowers TDD discipline and autonomous work patterns Memory, pre-mortems, validation across repeated sessions
Ruflo / Claude-Flow High-scale swarm orchestration and MCP-heavy coordination Local, auditable compounding around whatever executes the work

Detailed comparisons · Competitive radar


Docs

Topic Where
Published site boshu2.github.io/agentops
Start navigating Docs index
New contributor orientation Newcomer guide
Working with .agents/ Operator guide
Full skill catalog Skills
CLI reference CLI commands
Architecture Architecture
Behavioral discipline Behavior guide
FAQ FAQ

Building docs locally. The site is built with MkDocs Material. Python 3.10+ is required; the dev toolchain is pinned in requirements-docs.txt.

scripts/docs-build.sh --serve    # live-reload dev server at http://127.0.0.1:8000
scripts/docs-build.sh --check    # strict build (mirrors what CI runs)
scripts/docs-build.sh            # build site to _site/

The first run creates .venv-docs/ and installs the toolchain via uv (preferred) or pip. The deploy workflow at .github/workflows/docs.yml runs the same mkdocs build --strict on every push to main and publishes to GitHub Pages.


The 12-Factor Doctrine

AgentOps is shaped by a set of public principles — the 12 factors of agent operations. Foundation, Flow, Knowledge, and Scale. Read them at 12factoragentops.com.

Tier Factors
Foundation (I-III) Context Is Everything · Track Everything in Git · One Agent, One Job
Flow (IV-VI) Research Before You Build · Validate Externally · Lock Progress Forward
Knowledge (VII-IX) Extract Learnings · Compound Knowledge · Measure What Matters
Scale (X-XII) Isolate Workers · Supervise Hierarchically · Harvest Failures as Wisdom

The AgentOps product implements these principles through skills, the ao CLI, and local bookkeeping in .agents/. See each factor page at 12factoragentops.com/factors for the doctrine behind the mechanism.


Ready?

# 1. Install (pick your runtime above)
# 2. Run in your repo
ao quick-start
#    or: ao quickstart
# 3. Validate from your agent chat
/council validate this PR

Then explore the skills catalog, the ao CLI reference, and the 12-factor doctrine.


Contributing

See docs/CONTRIBUTING.md. Agent contributors should also read AGENTS.md and use bd for issue tracking.

License

Apache-2.0 · Docs · CLI Reference