AgentFeedback

Asked to install this? Read AGENT-INSTALL.md (client) or AGENT-INSTALL-STACK.md (server and client). Both describe release 4.0.0 and later.

Paste one of these into a coding agent:

Install the AgentFeedback client from https://github.com/AgentFeedback/agentfeedback and set it up for this harness.

Install the AgentFeedback stack from https://github.com/AgentFeedback/agentfeedback: server and client, and verify reporting end to end.

An inbox for feedback from AI coding agents. Agents report what slowed them down (a missing or wrong doc, a tool that misbehaved, stale config); later an agent triages the reports with a human and fixes the causes. Local first: one binary keeps the reports in a SQLite database on your machine, with no server and no key. A server is optional, for collecting reports from several machines.

Start on this machine

Install the agentfeedback binary with the checksum-verified command of AGENT-INSTALL.md step 2.1, then:

agentfeedback init

init asks two questions: where reports go (Enter for local) and which of the detected coding-agent harnesses to wire (Enter for all). It installs the agentfeedback skill, a one-line rule in each harness's global instruction file and hooks that suggest a report when tool calls keep failing, files and marks one test report, and prints what it did. agentfeedback init --yes asks nothing and takes those defaults. Then:

agentfeedback ui                   # browse the reports in a read-only page on this machine
agentfeedback list --open          # the reports not yet processed
agentfeedback uninstall all        # remove what init wired

Triage the reports with the agentfeedback-triage skill (init prints the command that installs it): invoke /agentfeedback-triage in the harness. To report to a server instead, run agentfeedback init --server <URL>; to run one, follow AGENT-INSTALL-STACK.md or docs/operate.md.

This is the open-source, self-hostable AgentFeedback (agentfeedback.dev). A hosted version runs at agentfeedback.io; it speaks the same API.

Start here

You want to	Read
Have an agent install the client, or the server and the client	AGENT-INSTALL.md, AGENT-INSTALL-STACK.md (from release 4.0.0)
Point an agent at every machine-readable document (llms.txt)	llms.txt
See what an agent is taught to file and how	skills/agentfeedback/SKILL.md (installed by agentfeedback init)
Triage the reports	skills/agentfeedback-triage/SKILL.md
Turn past coding-agent sessions on this machine into feedback (agentfeedback sessions)	docs/sessions.md
Call the HTTP API of a server	docs/api.md, the contract docs/openapi.yaml
File a report over HTTP without the binary, from a shell	docs/recipes/http-curl.md, docs/recipes/http-powershell.md
Give an agent the reference docs, schemas and OpenAPI document (integrate, operate)	skills/agentfeedback-docs/SKILL.md (installed by agentfeedback install --docs)
Wire harnesses, run a server, back up, migrate	docs/operate.md
Uninstall the skills or the service	docs/operate.md#uninstall
Change the code	AGENTS.md then docs/develop.md
Trust boundary and credentials	docs/security.md
Versions and upgrade notes	docs/releases.md
Submit AgentFeedback to an agent platform's catalog (maintainers)	docs/integrations/clawhub.md, docs/integrations/hermes.md, docs/integrations/nanoclaw.md

Where to get it

Besides this repository, these are the only official sources of AgentFeedback. A package of that name anywhere else, including the unrelated agent-feedback on npm and PyPI, does not come from this project.

Channel	Name	Status
GitHub Releases	AgentFeedback/agentfeedback releases, assets in docs/releases.md	from release 4.0.0
Container image	ghcr.io/agentfeedback/agentfeedback, see docs/operate.md	from release 4.0.0
Go module	github.com/agentfeedback/agentfeedback/v4, built with go install (docs/operate.md)	from release 4.0.0
Homebrew	tap agentfeedback/tap (repository AgentFeedback/homebrew-tap)	planned
npm	@agentfeedback/cli, the @agentfeedback scope	launcher 0.0.x; the binary is planned
PyPI	agentfeedback-cli	launcher 0.0.x; the binary is planned
RubyGems	agentfeedback	launcher 0.0.x; the binary is planned

Until they ship the binary, the npm, PyPI and RubyGems packages at 0.0.x are only a launcher: their agentfeedback command runs an agentfeedback binary already on PATH and otherwise prints the link to AGENT-INSTALL.md step 2.1.

Integrations

The table is generated from integrations.json; the status is how far each integration was checked, and when.

Integration	Kind	Install	Status
Claude Code	adapter	agentfeedback install claude-code	live-checked 2026-10-07
Codex CLI	adapter	agentfeedback install codex	documented 2026-10-07
Cursor	adapter	agentfeedback install cursor	documented 2026-10-07
OpenCode	adapter	agentfeedback install opencode	live-checked 2026-10-07
Oh My Pi (omp)	adapter	agentfeedback install omp	documented 2026-10-07
Pi	adapter	agentfeedback install pi	documented 2026-10-07
Copilot CLI	adapter	agentfeedback install copilot	documented 2026-10-07
Antigravity	adapter	agentfeedback install antigravity	documented 2026-10-07
Devin	adapter	agentfeedback install devin	documented 2026-10-07
Kiro	adapter	agentfeedback install kiro	documented 2026-10-07
Cline	adapter	agentfeedback install cline	documented 2026-10-07
Amp	adapter	agentfeedback install amp	documented 2026-10-07
VS Code	adapter	agentfeedback install vscode --mcp	documented 2026-10-07
Gemini CLI	adapter	agentfeedback install gemini-cli	documented 2026-10-07
OpenClaw	platform	the agentfeedback skill from ClawHub, once listed	documented 2026-10-09
Hermes Agent	platform	hermes plugins install agentfeedback, once listed	documented 2026-10-09
NanoClaw	platform	ncl groups create --template engineering/agentfeedback, once listed	documented 2026-10-09
skills.sh (npx skills)	hub	npx --yes skills add AgentFeedback/agentfeedback --skill agentfeedback -g -a <agent> -y	documented 2026-10-09
Claude Code plugin marketplace	hub	claude plugin marketplace add AgentFeedback/agentfeedback && claude plugin install agentfeedback@agentfeedback	documented 2026-10-09
Codex plugin marketplace	hub	codex plugin marketplace add AgentFeedback/agentfeedback && codex plugin add agentfeedback@agentfeedback	documented 2026-10-09

What it is not

Not a review runner, benchmark, dashboard or automated fixer. It stores what agents report and lets a processor work through it. Records are never overwritten: a correction is a new submission.

MIT licensed.