thumbgate 1.4.0 → 1.4.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.claude-plugin/README.md +25 -0
- package/.claude-plugin/marketplace.json +1 -1
- package/README.md +195 -168
- package/adapters/chatgpt/INSTALL.md +59 -4
- package/bin/cli.js +4 -0
- package/config/github-about.json +1 -1
- package/package.json +9 -5
- package/public/index.html +44 -23
- package/scripts/auto-promote-gates.js +5 -3
- package/scripts/billing-setup.js +109 -0
- package/scripts/build-claude-mcpb.js +71 -5
- package/scripts/distribution-surfaces.js +28 -0
- package/scripts/feedback-to-rules.js +27 -8
- package/scripts/gates-engine.js +51 -7
- package/scripts/hosted-config.js +2 -0
- package/scripts/hybrid-feedback-context.js +26 -16
- package/scripts/lesson-inference.js +20 -1
- package/scripts/operational-summary.js +41 -5
- package/scripts/ralph-loop.js +376 -0
- package/scripts/ralph-mode-ci.js +331 -0
- package/scripts/rotate-stripe-webhook-secret.js +314 -0
- package/scripts/statusline.sh +7 -2
- package/src/api/server.js +23 -3
- package/scripts/__pycache__/train_from_feedback.cpython-312.pyc +0 -0
package/.claude-plugin/README.md
CHANGED
|
@@ -38,12 +38,36 @@ https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-c
|
|
|
38
38
|
|
|
39
39
|
That bundle is built from the same `.claude-plugin` metadata in this repo and is meant for people who want a ready-to-install artifact instead of building locally.
|
|
40
40
|
|
|
41
|
+
### Review packet zip
|
|
42
|
+
|
|
43
|
+
Anthropic's submission flow may ask for a GitHub link or a zip that preserves the plugin folder structure. The review-ready source zip lives on GitHub Releases:
|
|
44
|
+
|
|
45
|
+
https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-claude-plugin-review.zip
|
|
46
|
+
|
|
41
47
|
### Anthropic directory path
|
|
42
48
|
|
|
43
49
|
If Anthropic approves the listing, install from Claude Desktop via `Settings -> Extensions`.
|
|
44
50
|
|
|
45
51
|
Directory inclusion is an external review process. Do not claim listing or approval before it is real.
|
|
46
52
|
|
|
53
|
+
Submission forms:
|
|
54
|
+
|
|
55
|
+
- https://claude.ai/settings/plugins/submit
|
|
56
|
+
- https://platform.claude.com/plugins/submit
|
|
57
|
+
|
|
58
|
+
### Repo marketplace while review is pending
|
|
59
|
+
|
|
60
|
+
Claude Code users do not need to wait for the official directory. Anthropic's plugin docs allow adding a repository marketplace directly when the repo contains `.claude-plugin/marketplace.json`.
|
|
61
|
+
|
|
62
|
+
Inside Claude Code, run:
|
|
63
|
+
|
|
64
|
+
```text
|
|
65
|
+
/plugin marketplace add IgorGanapolsky/ThumbGate
|
|
66
|
+
/plugin install thumbgate@thumbgate-marketplace
|
|
67
|
+
```
|
|
68
|
+
|
|
69
|
+
That uses the marketplace metadata already published in this repository while Anthropic reviews the official directory submission.
|
|
70
|
+
|
|
47
71
|
### MCPB bundle build
|
|
48
72
|
|
|
49
73
|
Maintainers can build the local Claude Desktop bundle directly from this repo:
|
|
@@ -131,4 +155,5 @@ For complete privacy information, see: https://thumbgate-production.up.railway.a
|
|
|
131
155
|
|
|
132
156
|
- Local Claude metadata lives in `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json`.
|
|
133
157
|
- The MCPB bundle is built with `npm run build:claude-mcpb`.
|
|
158
|
+
- The review packet zip is built with `npm run build:claude-review-zip`.
|
|
134
159
|
- Anthropic directory requirements and the internal publish checklist live in `docs/CLAUDE_DESKTOP_EXTENSION.md`.
|
package/README.md
CHANGED
|
@@ -1,164 +1,128 @@
|
|
|
1
1
|
# ThumbGate
|
|
2
2
|
|
|
3
|
-
|
|
3
|
+
**Thumbs up or thumbs down — and your AI coding agent never makes the same mistake twice.**
|
|
4
4
|
|
|
5
5
|
[](https://github.com/IgorGanapolsky/ThumbGate/actions/workflows/ci.yml)
|
|
6
6
|
[](https://www.npmjs.com/package/thumbgate)
|
|
7
7
|
[](LICENSE)
|
|
8
8
|
[](https://thumbgate-production.up.railway.app/?utm_source=github&utm_medium=readme&utm_campaign=badge_cta#workflow-sprint-intake)
|
|
9
9
|
|
|
10
|
-
**[Workflow Hardening Sprint](https://thumbgate-production.up.railway.app/?utm_source=github&utm_medium=readme&utm_campaign=top_cta#workflow-sprint-intake)** · **[
|
|
10
|
+
**[Workflow Hardening Sprint](https://thumbgate-production.up.railway.app/?utm_source=github&utm_medium=readme&utm_campaign=top_cta#workflow-sprint-intake)** · **[Use in ChatGPT](adapters/chatgpt/INSTALL.md)** · **[Install Claude Desktop Extension](https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-claude-desktop.mcpb)** · **[Claude Plugin Guide](docs/CLAUDE_DESKTOP_EXTENSION.md)** · **[Install Codex Plugin](https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-codex-plugin.zip)** · **[Live Dashboard](https://thumbgate-production.up.railway.app/dashboard?utm_source=github&utm_medium=readme&utm_campaign=top_cta)** · **[Pro Page](https://thumbgate-production.up.railway.app/pro?utm_source=github&utm_medium=readme&utm_campaign=pro_page)**
|
|
11
11
|
|
|
12
|
-
**Popular buyer questions:** **[
|
|
12
|
+
**Popular buyer questions:** **[Stop repeated AI agent mistakes](https://thumbgate-production.up.railway.app/guides/stop-repeated-ai-agent-mistakes?utm_source=github&utm_medium=readme&utm_campaign=buyer_questions)** · **[Cursor guardrails](https://thumbgate-production.up.railway.app/guides/cursor-agent-guardrails?utm_source=github&utm_medium=readme&utm_campaign=buyer_questions)** · **[Codex CLI guardrails](https://thumbgate-production.up.railway.app/guides/codex-cli-guardrails?utm_source=github&utm_medium=readme&utm_campaign=buyer_questions)** · **[Gemini CLI memory + enforcement](https://thumbgate-production.up.railway.app/guides/gemini-cli-feedback-memory?utm_source=github&utm_medium=readme&utm_campaign=buyer_questions)**
|
|
13
13
|
|
|
14
|
-
**Running
|
|
14
|
+
**Running Claude Desktop?** **[Download Claude bundle](https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-claude-desktop.mcpb)** · **[Install + submission guide](docs/CLAUDE_DESKTOP_EXTENSION.md)** · **[Review packet zip](https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-claude-plugin-review.zip)**
|
|
15
15
|
|
|
16
|
-
|
|
16
|
+
**Running Codex?** **[Download the standalone Codex plugin bundle](https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-codex-plugin.zip)** · **[Codex install guide](plugins/codex-profile/INSTALL.md)**
|
|
17
17
|
|
|
18
|
-
**
|
|
18
|
+
**Running ChatGPT?** Search GPT Store for `ThumbGate` by Igor Ganapolsky. Regular users reply with 👍/👎 or "thumbs up/down" on answers, save the lesson, prevent repeated bad answers, and reinforce the answers that worked. Developers can import the prepared **[GPT Actions OpenAPI spec](adapters/chatgpt/openapi.yaml)** with the **[ChatGPT install guide](adapters/chatgpt/INSTALL.md)**.
|
|
19
19
|
|
|
20
|
-
|
|
20
|
+
**Official directory pending review?** Claude Code users can install today with `/plugin marketplace add IgorGanapolsky/ThumbGate` then `/plugin install thumbgate@thumbgate-marketplace`.
|
|
21
21
|
|
|
22
|
-
|
|
22
|
+
---
|
|
23
23
|
|
|
24
|
-
|
|
24
|
+
## What problem does this solve?
|
|
25
25
|
|
|
26
|
-
|
|
26
|
+
AI agents repeat mistakes. You fix the same problem in session after session — force-push to main, broken migrations, unauthorized file edits — because the agent has no memory of your feedback.
|
|
27
27
|
|
|
28
|
-
|
|
29
|
-
|
|
28
|
+
```
|
|
29
|
+
┌─────────────────────────────────────────────────────────────┐
|
|
30
|
+
│ THE PROBLEM │
|
|
31
|
+
│ │
|
|
32
|
+
│ Session 1: Agent breaks something. You fix it. │
|
|
33
|
+
│ Session 2: Agent breaks it again. You fix it again. │
|
|
34
|
+
│ Session 3: Same thing. Again. │
|
|
35
|
+
│ │
|
|
36
|
+
│ THE SOLUTION │
|
|
37
|
+
│ │
|
|
38
|
+
│ Session 1: Agent breaks something. You 👎 it. │
|
|
39
|
+
│ Session 2: ⛔ Gate blocks the mistake before it happens. │
|
|
40
|
+
│ Session 3+: Never see it again. │
|
|
41
|
+
└─────────────────────────────────────────────────────────────┘
|
|
42
|
+
```
|
|
30
43
|
|
|
31
|
-
|
|
44
|
+
ThumbGate is the **control plane** for AI coding agents — turning your feedback into **enforced rules**, not suggestions.
|
|
32
45
|
|
|
33
|
-
|
|
46
|
+
---
|
|
34
47
|
|
|
35
|
-
|
|
48
|
+
## How It Works in 3 Steps
|
|
36
49
|
|
|
37
|
-
```bash
|
|
38
|
-
npx thumbgate init
|
|
39
50
|
```
|
|
51
|
+
STEP 1 STEP 2 STEP 3
|
|
52
|
+
──────── ──────── ────────
|
|
40
53
|
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
ThumbGate is the control plane for AI coding agents:
|
|
44
|
-
|
|
45
|
-
- Feedback becomes enforcement, so repeated failures stop at the gate instead of reappearing in review.
|
|
46
|
-
- Workflow Sentinel scores blast radius before execution, so risky PR, release, and publish flows are visible early.
|
|
47
|
-
- High-risk local actions can be routed into Docker Sandboxes, while hosted team automations use a signed isolated sandbox lane.
|
|
48
|
-
- Team rollout stays tied to [Verification Evidence](docs/VERIFICATION_EVIDENCE.md) instead of trust-me operator claims.
|
|
49
|
-
- AI agent outputs stay grounded in your team's actual standards — not generic patterns — because every gate enforces human judgment before the action executes.
|
|
50
|
-
|
|
51
|
-
## Release Confidence
|
|
54
|
+
You react ThumbGate learns The gate holds
|
|
52
55
|
|
|
53
|
-
|
|
56
|
+
👎 on a bad ──► Feedback becomes ──► Next time the
|
|
57
|
+
agent action a saved lesson agent tries the
|
|
58
|
+
and a block rule same thing:
|
|
59
|
+
👍 on a good ──► Good pattern gets ⛔ BLOCKED
|
|
60
|
+
agent action reinforced (or ✅ allowed)
|
|
61
|
+
```
|
|
54
62
|
|
|
55
|
-
|
|
56
|
-
- [SemVer Policy](docs/SEMVER_POLICY.md) and version-sync checks keep `package.json`, `CHANGELOG.md`, plugin manifests, and installer metadata aligned.
|
|
57
|
-
- CI enforces changeset coverage, version sync, tests, coverage, proof lanes, and operational integrity before merge.
|
|
58
|
-
- Final close-out requires verifying the exact `main` merge commit, with proof anchored in [Verification Evidence](docs/VERIFICATION_EVIDENCE.md).
|
|
63
|
+
That's it. No manual rule-writing. No config files to maintain. Your reactions teach the agent what your team actually wants.
|
|
59
64
|
|
|
60
|
-
|
|
65
|
+
---
|
|
61
66
|
|
|
62
67
|
## Before / After
|
|
63
68
|
|
|
64
69
|
```
|
|
65
|
-
WITHOUT THUMBGATE
|
|
66
|
-
|
|
67
|
-
Session 1:
|
|
68
|
-
Agent force-pushes to main.
|
|
69
|
-
You correct it.
|
|
70
|
-
|
|
71
|
-
Session 2:
|
|
72
|
-
Agent force-pushes again.
|
|
73
|
-
It learned nothing.
|
|
74
|
-
|
|
75
|
-
Session 3:
|
|
76
|
-
Same mistake. Again.
|
|
70
|
+
WITHOUT THUMBGATE │ WITH THUMBGATE
|
|
71
|
+
───────────────────────────────┼───────────────────────────────
|
|
72
|
+
Session 1: │ Session 1:
|
|
73
|
+
Agent force-pushes to main. │ Agent force-pushes to main.
|
|
74
|
+
You correct it manually. │ You 👎 it.
|
|
75
|
+
│
|
|
76
|
+
Session 2: │ Session 2:
|
|
77
|
+
Agent force-pushes again. │ ⛔ Gate blocks force-push.
|
|
78
|
+
It learned nothing. │ Agent uses safe push instead.
|
|
79
|
+
│
|
|
80
|
+
Session 3: │ Session 3+:
|
|
81
|
+
Same mistake. Again. │ Permanently fixed.
|
|
82
|
+
And again. │
|
|
77
83
|
```
|
|
78
84
|
|
|
79
|
-
|
|
85
|
+
---
|
|
86
|
+
|
|
87
|
+
## The Feedback Loop
|
|
80
88
|
|
|
81
89
|
```
|
|
82
|
-
|
|
83
|
-
│
|
|
84
|
-
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
│ │ └─────────────────┘ │
|
|
91
|
-
│ │ │
|
|
92
|
-
│ │ PreToolUse hook fires │
|
|
93
|
-
│ │◄───────────────────────────┤ tries same mistake
|
|
94
|
-
│ │ ⛔ BLOCKED │
|
|
95
|
-
│ ├───────────────────────────►│ forced to try safe path
|
|
96
|
-
│ │ │
|
|
97
|
-
│ 👍 "good fix" │ │
|
|
98
|
-
├───────────────────────►│ │
|
|
99
|
-
│ │ reinforced ✅ │
|
|
100
|
-
│ │ │
|
|
90
|
+
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
|
|
91
|
+
│ Capture │───►│ Learn │───►│ Remember │───►│ Rule │───►│ Gate │
|
|
92
|
+
│ │ │ │ │ │ │ │ │ │
|
|
93
|
+
│ 👍 / 👎 │ │ Feedback │ │ Stored │ │ Auto- │ │ Blocks │
|
|
94
|
+
│ │ │ becomes │ │ lessons │ │ generated│ │ bad │
|
|
95
|
+
│ │ │ a lesson │ │ & search │ │ from │ │ actions │
|
|
96
|
+
│ │ │ │ │ │ │ feedback │ │ live │
|
|
97
|
+
└──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘
|
|
101
98
|
```
|
|
102
99
|
|
|
103
|
-
|
|
104
|
-
|
|
105
|
-
- **Stop AI agent force-push to main** — Prevent lost commits with a pre-action gate that blocks `git push --force` on protected branches
|
|
106
|
-
- **Prevent repeated database migration failures** — Each mistake becomes a searchable lesson that fires before the next migration attempt
|
|
107
|
-
- **Block unauthorized file edits** — Control which files agents can modify with path-based gates
|
|
108
|
-
- **Memory across sessions** — Agent remembers feedback from yesterday's mistakes without any manual rule-writing
|
|
109
|
-
- **Shared team safety** — One developer's thumbs-down protects the whole team from the same mistake
|
|
110
|
-
- **Auto-improving without human feedback** — Self-distillation mode evaluates agent outcomes and generates lessons automatically
|
|
111
|
-
|
|
112
|
-
## FAQ
|
|
113
|
-
|
|
114
|
-
**Is ThumbGate a model fine-tuning tool?**
|
|
115
|
-
No. ThumbGate doesn't update model weights. It works by capturing feedback into structured lessons, injecting relevant context at runtime, and blocking bad actions via PreToolUse hooks.
|
|
116
|
-
|
|
117
|
-
**How is this different from CLAUDE.md or .cursorrules?**
|
|
118
|
-
CLAUDE.md files are suggestions that agents can ignore. ThumbGate gates are enforcement — they physically block the action before it executes via PreToolUse hooks. Gates also auto-generate from feedback instead of requiring manual rule-writing.
|
|
100
|
+
---
|
|
119
101
|
|
|
120
|
-
|
|
121
|
-
Yes. ThumbGate is MCP-compatible and works with Claude Code, Cursor, Codex, Gemini CLI, Amp, OpenCode, and any agent that supports PreToolUse hooks or MCP. Codex now has a standalone plugin bundle at `https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-codex-plugin.zip` in addition to the repo-local profile.
|
|
102
|
+
## Get Started
|
|
122
103
|
|
|
123
|
-
**
|
|
124
|
-
ThumbGate can auto-evaluate agent action outcomes (test failures, reverted edits, error patterns) and generate prevention rules without any human feedback. Your agent gets smarter every session automatically.
|
|
104
|
+
**Best first paid motion for teams:** the **Workflow Hardening Sprint** — qualify one repeated failure before committing to a full rollout. **[Start intake →](https://thumbgate-production.up.railway.app/?utm_source=github&utm_medium=readme&utm_campaign=team_rollout#workflow-sprint-intake)**
|
|
125
105
|
|
|
126
|
-
**
|
|
127
|
-
Free tier: 3 feedback captures/day, 5 lesson searches/day, 5 built-in gates. Pro is $19/mo or $149/yr for solo operators who need the personal local dashboard and exports. Team rollout starts intake-first at $99/seat/mo with a 3-seat minimum when shared lessons, org visibility, and approval boundaries matter.
|
|
106
|
+
**Best first technical motion:** install the CLI-first and let `init` wire hooks for the agent you already use.
|
|
128
107
|
|
|
129
|
-
|
|
130
|
-
|
|
131
|
-
```
|
|
132
|
-
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
|
|
133
|
-
│ Capture │────►│ Distill │────►│ Remember │────►│ Rule │────►│ Gate │
|
|
134
|
-
│ 👍 / 👎 │ │ history- │ │ SQLite + │ │ auto-gen │ │ PreTool │
|
|
135
|
-
│ │ │ aware │ │ FTS5 DB │ │ from │ │ Use hook │
|
|
136
|
-
│ │ │ │ │ │ │ failures │ │ enforces │
|
|
137
|
-
└──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘
|
|
138
|
-
```
|
|
108
|
+
**Paid path for individual operators:** [ThumbGate Pro](https://thumbgate-production.up.railway.app/pro?utm_source=github&utm_medium=readme&utm_campaign=pro_page) is the self-serve side lane for a personal dashboard and export-ready evidence.
|
|
139
109
|
|
|
140
|
-
|
|
110
|
+
---
|
|
141
111
|
|
|
142
|
-
|
|
112
|
+
## Quick Start
|
|
143
113
|
|
|
144
114
|
```bash
|
|
145
|
-
npx thumbgate init
|
|
146
|
-
npx thumbgate doctor
|
|
147
|
-
npx thumbgate lessons
|
|
148
|
-
npx thumbgate dashboard
|
|
115
|
+
npx thumbgate init # detects your agent and wires everything up
|
|
116
|
+
npx thumbgate doctor # health check
|
|
117
|
+
npx thumbgate lessons # see what's been learned
|
|
118
|
+
npx thumbgate dashboard # open local dashboard
|
|
149
119
|
```
|
|
150
120
|
|
|
151
|
-
Or wire MCP directly: `claude mcp add thumbgate -- npx
|
|
152
|
-
|
|
153
|
-
Works with **Claude Code, Cursor, Codex, Gemini, Amp, OpenCode**, and any MCP-compatible agent.
|
|
154
|
-
|
|
155
|
-
Codex standalone plugin bundle: `https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-codex-plugin.zip`
|
|
121
|
+
Or wire MCP directly: `claude mcp add thumbgate -- npx --yes --package thumbgate thumbgate serve`
|
|
156
122
|
|
|
157
|
-
Codex
|
|
123
|
+
Works with **Claude Code, Cursor, Codex, Gemini CLI, Amp, OpenCode**, and any MCP-compatible agent.
|
|
158
124
|
|
|
159
|
-
|
|
160
|
-
>
|
|
161
|
-
> **Need a personal dashboard and DPO export for yourself?** [See ThumbGate Pro →](https://thumbgate-production.up.railway.app/pro?utm_source=github&utm_medium=readme&utm_campaign=quickstart_cta_pro)
|
|
125
|
+
---
|
|
162
126
|
|
|
163
127
|
## Install for Your Agent
|
|
164
128
|
|
|
@@ -166,13 +130,13 @@ Codex install guide: `plugins/codex-profile/INSTALL.md`
|
|
|
166
130
|
```bash
|
|
167
131
|
npx thumbgate init --agent claude-code
|
|
168
132
|
```
|
|
169
|
-
Wires
|
|
133
|
+
Wires hooks automatically. Works immediately.
|
|
170
134
|
|
|
171
135
|
### Cursor
|
|
172
136
|
```bash
|
|
173
137
|
npx thumbgate init --agent cursor
|
|
174
138
|
```
|
|
175
|
-
Installs as a Cursor extension with 4 skills: capture
|
|
139
|
+
Installs as a Cursor extension with 4 skills: capture feedback, manage rules, search lessons, recall context.
|
|
176
140
|
|
|
177
141
|
### Codex
|
|
178
142
|
```bash
|
|
@@ -203,100 +167,163 @@ Add to your `claude_desktop_config.json`:
|
|
|
203
167
|
"mcpServers": {
|
|
204
168
|
"thumbgate": {
|
|
205
169
|
"command": "npx",
|
|
206
|
-
"args": ["--yes", "thumbgate", "serve"]
|
|
170
|
+
"args": ["--yes", "--package", "thumbgate", "thumbgate", "serve"]
|
|
207
171
|
}
|
|
208
172
|
}
|
|
209
173
|
}
|
|
210
174
|
```
|
|
175
|
+
Or [download the packaged extension bundle](https://github.com/IgorGanapolsky/ThumbGate/releases/latest/download/thumbgate-claude-desktop.mcpb) and install directly.
|
|
211
176
|
|
|
212
|
-
|
|
177
|
+
---
|
|
213
178
|
|
|
214
|
-
|
|
215
|
-
|
|
216
|
-
|
|
217
|
-
|
|
218
|
-
|
|
219
|
-
|
|
220
|
-
|
|
221
|
-
|
|
222
|
-
|
|
223
|
-
|
|
224
|
-
│ + custom gates in config/gates/custom.json │
|
|
225
|
-
└─────────────────────────────────────────────────────────┘
|
|
226
|
-
```
|
|
179
|
+
## Use Cases
|
|
180
|
+
|
|
181
|
+
- **Stop force-push to main** — A gate blocks `git push --force` on protected branches before it runs
|
|
182
|
+
- **Prevent repeated migration failures** — Each mistake becomes a searchable lesson that fires before the next attempt
|
|
183
|
+
- **Block unauthorized file edits** — Control which files agents can touch with path-based rules
|
|
184
|
+
- **Memory across sessions** — The agent remembers your feedback from yesterday without any manual rule-writing
|
|
185
|
+
- **Shared team safety** — One developer's thumbs-down protects the whole team from the same mistake
|
|
186
|
+
- **Auto-improving without feedback** — Self-improvement mode evaluates outcomes and generates rules automatically
|
|
187
|
+
|
|
188
|
+
---
|
|
227
189
|
|
|
228
190
|
## Feedback Sessions
|
|
229
191
|
|
|
192
|
+
Give the agent more context when a thumbs-down isn't enough:
|
|
193
|
+
|
|
230
194
|
```
|
|
231
195
|
👎 thumbs down
|
|
232
196
|
└─► open_feedback_session
|
|
233
|
-
└─► "you lied about deployment"
|
|
234
|
-
└─► "tests were actually failing"
|
|
197
|
+
└─► "you lied about deployment" (append_feedback_context)
|
|
198
|
+
└─► "tests were actually failing" (append_feedback_context)
|
|
235
199
|
└─► finalize_feedback_session
|
|
236
200
|
└─► lesson inferred from full conversation
|
|
237
201
|
```
|
|
238
202
|
|
|
239
|
-
|
|
203
|
+
ThumbGate uses up to 8 prior conversation entries to turn vague, history-aware negative signals into specific, actionable lessons. A 60-second follow-up window stays open for additional context via `open_feedback_session` → `append_feedback_context` → `finalize_feedback_session`.
|
|
240
204
|
|
|
241
205
|
Free and self-hosted users can invoke `search_lessons` directly through MCP, and via the CLI with `npx thumbgate lessons`.
|
|
242
206
|
|
|
243
|
-
|
|
207
|
+
---
|
|
208
|
+
|
|
209
|
+
## Built-in Gates
|
|
244
210
|
|
|
245
211
|
```
|
|
246
|
-
|
|
247
|
-
│
|
|
248
|
-
|
|
249
|
-
│
|
|
250
|
-
│
|
|
251
|
-
│
|
|
252
|
-
│
|
|
253
|
-
│
|
|
254
|
-
│
|
|
255
|
-
|
|
212
|
+
┌─────────────────────────────────────────────────────────┐
|
|
213
|
+
│ ENFORCEMENT LAYER │
|
|
214
|
+
│ │
|
|
215
|
+
│ ⛔ force-push → blocks git push --force │
|
|
216
|
+
│ ⛔ protected-branch → blocks direct push to main │
|
|
217
|
+
│ ⛔ unresolved-threads → blocks push with open reviews │
|
|
218
|
+
│ ⛔ package-lock-reset → blocks destructive lock edits │
|
|
219
|
+
│ ⛔ env-file-edit → blocks .env secret exposure │
|
|
220
|
+
│ │
|
|
221
|
+
│ + custom gates in config/gates/custom.json │
|
|
222
|
+
└─────────────────────────────────────────────────────────┘
|
|
256
223
|
```
|
|
257
224
|
|
|
258
|
-
|
|
225
|
+
---
|
|
226
|
+
|
|
227
|
+
## Pricing
|
|
228
|
+
|
|
229
|
+
```
|
|
230
|
+
┌──────────────────┬──────────────────────────────┬──────────────────────┐
|
|
231
|
+
│ FREE │ TEAM $99/seat/mo (min 3) │ PRO $19/mo · $149/yr│
|
|
232
|
+
├──────────────────┼──────────────────────────────┼──────────────────────┤
|
|
233
|
+
│ Local CLI │ Workflow Hardening Sprint │ Personal dashboard │
|
|
234
|
+
│ Enforced gates │ Shared hosted lesson DB │ Export feedback data │
|
|
235
|
+
│ 3 captures/day │ Org-wide dashboard │ Review-ready exports │
|
|
236
|
+
│ 5 searches/day │ Approval + audit proof │ │
|
|
237
|
+
│ Unlimited recall │ Isolated execution guidance │ │
|
|
238
|
+
└──────────────────┴──────────────────────────────┴──────────────────────┘
|
|
239
|
+
```
|
|
259
240
|
|
|
260
|
-
|
|
241
|
+
**[Start Workflow Hardening Sprint](https://thumbgate-production.up.railway.app/?utm_source=github&utm_medium=readme&utm_campaign=top_cta#workflow-sprint-intake)** · **[Live Dashboard](https://thumbgate-production.up.railway.app/dashboard?utm_source=github&utm_medium=readme&utm_campaign=top_cta)** · **[See Pro](https://thumbgate-production.up.railway.app/pro?utm_source=github&utm_medium=readme&utm_campaign=pro_page)**
|
|
261
242
|
|
|
262
|
-
|
|
243
|
+
**Where to start:**
|
|
244
|
+
- **Teams:** Begin with the Workflow Hardening Sprint — qualify one real repeated failure before committing to a full rollout
|
|
245
|
+
- **Solo operators:** ThumbGate Pro adds a personal dashboard and export-ready evidence
|
|
246
|
+
- **Individuals & open source:** Free CLI tier, self-hosted
|
|
263
247
|
|
|
264
|
-
|
|
248
|
+
---
|
|
265
249
|
|
|
266
250
|
## Tech Stack
|
|
267
251
|
|
|
268
252
|
```
|
|
269
|
-
|
|
270
|
-
│
|
|
271
|
-
│
|
|
272
|
-
│
|
|
273
|
-
│
|
|
274
|
-
│
|
|
275
|
-
│
|
|
276
|
-
|
|
277
|
-
│
|
|
278
|
-
|
|
279
|
-
│
|
|
280
|
-
│
|
|
281
|
-
│
|
|
282
|
-
│
|
|
283
|
-
│
|
|
284
|
-
|
|
253
|
+
┌──────────────────────┬──────────────────────┬──────────────────────┐
|
|
254
|
+
│ STORAGE │ INTELLIGENCE │ ENFORCEMENT │
|
|
255
|
+
│ │ │ │
|
|
256
|
+
│ SQLite + FTS5 │ MemAlign dual recall │ PreToolUse hook │
|
|
257
|
+
│ LanceDB vectors │ Thompson Sampling │ engine │
|
|
258
|
+
│ JSONL logs │ (adaptive lesson │ Gates config │
|
|
259
|
+
│ File-based context │ selection) │ Hook wiring │
|
|
260
|
+
│ │ │ │
|
|
261
|
+
│ │ │ │
|
|
262
|
+
├──────────────────────┼──────────────────────┼──────────────────────┤
|
|
263
|
+
│ INTERFACES │ BILLING │ EXECUTION │
|
|
264
|
+
│ │ │ │
|
|
265
|
+
│ MCP stdio │ Stripe │ Railway │
|
|
266
|
+
│ HTTP API │ │ Cloudflare Workers │
|
|
267
|
+
│ CLI │ │ Docker Sandboxes │
|
|
268
|
+
│ Node.js >=18 │ │ │
|
|
269
|
+
└──────────────────────┴──────────────────────┴──────────────────────┘
|
|
285
270
|
```
|
|
286
271
|
|
|
272
|
+
---
|
|
273
|
+
|
|
274
|
+
## FAQ
|
|
275
|
+
|
|
276
|
+
**Is ThumbGate a model fine-tuning tool?**
|
|
277
|
+
No. ThumbGate does not update model weights in frontier LLMs. It captures your feedback, stores lessons, injects context at runtime, and blocks bad actions before they execute.
|
|
278
|
+
|
|
279
|
+
**How is this different from CLAUDE.md or .cursorrules?**
|
|
280
|
+
Those are suggestions the agent can ignore. ThumbGate gates are enforced — they physically block the action before it runs. They also auto-generate from feedback instead of requiring manual writing.
|
|
281
|
+
|
|
282
|
+
**Does it work with my agent?**
|
|
283
|
+
Yes. It's MCP-compatible and works with Claude Code, Claude Desktop, Cursor, Codex, Gemini CLI, Amp, OpenCode, and any agent that supports MCP or pre-action hooks.
|
|
284
|
+
|
|
285
|
+
**What's self-improvement mode?**
|
|
286
|
+
ThumbGate can watch for failure signals (test failures, reverted edits, error patterns) and auto-generate prevention rules — no thumbs-down required. Your agent gets smarter every session.
|
|
287
|
+
|
|
288
|
+
**Is it free?**
|
|
289
|
+
Free tier: **3 daily feedback captures**, **5 daily lesson searches**, unlimited recall, enforced gates. History-aware distillation turns vague feedback into specific lessons. Pro is $19/mo or $149/yr for a personal dashboard and exports. Team rollout starts at $99/seat/mo (3-seat minimum) with shared hosted lesson DB, org dashboard, approval + audit proof, and isolated execution guidance.
|
|
290
|
+
|
|
291
|
+
---
|
|
292
|
+
|
|
293
|
+
## Enterprise Story
|
|
294
|
+
|
|
295
|
+
ThumbGate is the control plane for AI coding agents:
|
|
296
|
+
|
|
297
|
+
- Feedback becomes enforcement — repeated failures stop at the gate instead of reappearing in review.
|
|
298
|
+
- **Workflow Sentinel** scores blast radius before execution, so risky PR, release, and publish flows are visible early.
|
|
299
|
+
- High-risk local actions route into **Docker Sandboxes**; hosted team automations use a signed isolated sandbox lane.
|
|
300
|
+
- Team rollout stays tied to [Verification Evidence](docs/VERIFICATION_EVIDENCE.md) instead of trust-me operator claims.
|
|
301
|
+
|
|
302
|
+
## Release Confidence
|
|
303
|
+
|
|
304
|
+
- Every PR must carry a **Changeset** entry — each shipped version has a customer-readable explanation before publish.
|
|
305
|
+
- Version-sync checks keep `package.json`, `CHANGELOG.md`, plugin manifests, and installer metadata aligned.
|
|
306
|
+
- Final close-out requires verifying the exact `main` merge commit, with proof anchored in [Verification Evidence](docs/VERIFICATION_EVIDENCE.md).
|
|
307
|
+
|
|
308
|
+
See [Release Confidence](docs/RELEASE_CONFIDENCE.md) for the full trust chain.
|
|
309
|
+
|
|
310
|
+
---
|
|
311
|
+
|
|
287
312
|
## Docs
|
|
288
313
|
|
|
289
314
|
- [Commercial Truth](docs/COMMERCIAL_TRUTH.md) — pricing, claims, what we don't say
|
|
290
|
-
- [Changeset Strategy](docs/CHANGESET_STRATEGY.md) — how release notes
|
|
291
|
-
- [First Dollar Playbook](docs/FIRST_DOLLAR_PLAYBOOK.md) —
|
|
292
|
-
- [Release Confidence](docs/RELEASE_CONFIDENCE.md) — how
|
|
315
|
+
- [Changeset Strategy](docs/CHANGESET_STRATEGY.md) — how release notes and version bumps are enforced
|
|
316
|
+
- [First Dollar Playbook](docs/FIRST_DOLLAR_PLAYBOOK.md) — turning one painful workflow into the next booked pilot
|
|
317
|
+
- [Release Confidence](docs/RELEASE_CONFIDENCE.md) — how changesets, version checks, and proof lanes make publishes inspectable
|
|
293
318
|
- [SemVer Policy](docs/SEMVER_POLICY.md) — stable vs prerelease channel rules
|
|
294
319
|
- [Verification Evidence](docs/VERIFICATION_EVIDENCE.md) — proof artifacts
|
|
295
320
|
- [WORKFLOW.md](WORKFLOW.md) — agent-run contract (scope, hard stops, proof commands)
|
|
296
|
-
- [
|
|
321
|
+
- [Ready-for-agent issue template](.github/ISSUE_TEMPLATE/ready-for-agent.yml) — intake for agent tasks
|
|
297
322
|
|
|
298
323
|
Pro overlay: [`thumbgate-pro`](https://github.com/IgorGanapolsky/thumbgate-pro) — separate repo/package inheriting from this base.
|
|
299
324
|
|
|
325
|
+
---
|
|
326
|
+
|
|
300
327
|
## License
|
|
301
328
|
|
|
302
329
|
MIT. See [LICENSE](LICENSE).
|