greenflash
Analyze real user-agent conversations to surface where users get blocked, optimize prompts, and understand what drives engagement or churn
Open source Repository Open in the app JSON README (API)
About
Analyze real user-agent conversations to surface where users get blocked, optimize prompts, and understand what drives engagement or churn
Details
- Kind
- Plugins
- Topic
- No topic detected
- Publisher
- greenflash-ai
- Origin
- marketplace
- Category
- ferramentas
- Open pull requests
- 1
- Last push
- 2026-06-18T20:36:27Z
- Repository state
- ativo
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
greenflash-ai/agent-skills/greenflash
README
# Greenflash Agent Skills
Greenflash analyzes real user-agent conversations to surface where users get blocked, which flows fail, and which interactions drive upgrades or churn. These skills bring that intelligence directly into your coding agent so you can ask questions, get answers, and apply prompt and tool fixes directly when Greenflash recommends them.
## Installation
### Any Agent (Skills CLI)
```bash
npx skills add greenflash-ai/agent-skills
```
### Claude Code (Marketplace)
Run these as **slash commands inside Claude Code** (not in your shell). Start `claude`, then:
```
/plugin marketplace add greenflash-ai/agent-skills
/plugin install greenflash@greenflash-plugins
```
A few things to know:
- These are slash commands inside the Claude Code REPL
- You'll see a "trust this repository" prompt the first time — accept it.
- Plugins activate on the next session, or run `/reload-plugins` to pick them up immediately.
No environment setup required. On first run, the skill activates a fresh API key for you via OAuth — it opens a browser tab, you click **Authorize**, and the key is saved to a gitignored `.greenflash` file in the project. If you can't open a browser (headless, remote SSH), you'll get a paste-key fallback automatically.
## Permissions
Each skill declares the exact tools it needs in its `allowed-tools` frontmatter (`curl` to the Greenflash API, reading/writing the `.greenflash` credentials file, opening the activation URL), so **normal use in Claude Code runs without permission prompts**.
Claude Code plugins can't grant permissions on your behalf, so for a totally prompt-free experience across every project — or when driving these skills from another agent — add the same allowlist to your `~/.claude/settings.json` (or a project `.claude/settings.json`):
```json
{
"permissions": {
"allow": [
"Bash(curl:*)",
"Bash(printenv GREENFLASH_API_KEY)",
"Bash(printenv GREENFLASH_API_URL)",
"Bash(cat .greenflash*)",
"Read(//**/.greenflash)",
"Write(//**/.greenflash)",
"Edit(//**/.greenflash)"
]
}
}
```
**Deep dives stay within this allowlist by design.** A broad investigation ("analyze my last 50 conversations") is answered by a single Chat API call, not by hand-pulling and parsing records — so it needs no extra permissions. When raw records are required, the skills fetch them with individual `curl` calls and inspect the saved output with the Read tool, deliberately avoiding the pipes, loops, and `python3`/`jq` pipelines that would otherwise prompt. See `skills/greenflash-config.md` → *Deep Dives & Bulk Investigation*.
## Skills
| Skill | Description |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------- |
| `greenflash` | Entry point. Auto-detects if Greenflash is set up, offers onboarding if not, and routes your question to the right workflow |
| `greenflash-health` | Surface quality trends, anomalies, safety issues, and sentiment across your products |
| `greenflash-inbox` | Triage flagged conversations, prioritized by severity |
| `greenflash-users` | Understand user behavior: who's struggling, who's churning, and what segments look like |
| `greenflash-prompts` | Find prompt and model quality issues, get optimization recommendations, and apply fixes |
| `greenflash-diagnose` | Root cause analysis for failing tools, user friction, and guardrail violations. Implements fixes directly |
| `greenflash-tickets` | Draft and file tickets in Linear (or your connected ticket provider) from conversation evidence, with a two-step draft → confirm flow |
| `greenflash-update-context` | Correct a Greenflash analysis you disagree with. Sends one specific correction, in the user's own words, to the product's context notes |
### SDK Integration
| Skill | Description |
| ---------------------------- | -------------------------------------------------------------------------------------------------------------------- |
| `greenflash-onboard-unified` | One-command setup: auto-detects your codebase and walks through all integration steps |
| `greenflash-verify` | Verify your integration is working: API key, SDK, client setup, and data flow |
| `greenflash-onboard` | Low-level: SDK install + conversation logging only. Most users should reach for `greenflash-onboard-unified` instead |
| `greenflash-onboard-prompts` | Log system prompts for automatic versioning and optimization |
| `greenflash-onboard-agentic` | Log structured message types for agent tool calls, reasoning traces, and workflows |
| `greenflash-onboard-events` | Track business events and link AI conversations to real outcomes like conversions and churn |
## Local Development
If you're running the Greenflash API locally, set `GREENFLASH_API_URL` to override the default production URL:
```bash
export GREENFLASH_API_URL=http://localhost:3000/api/v1
```
## Shared Config (contributors)
`skills/greenflash/greenflash-config.md` is the single canonical shared config (authentication, API patterns, error handling) — exactly one copy in the repo, no duplication, no symlinks. It lives inside the `greenflash` **entry skill** so it ships on every install path (the skills CLI installs each skill folder as-is, and all skills land as siblings under one directory).
Each skill reads it locally, no network:
- The `greenflash` entry skill reads its own co-located copy.
- The feature skills (`greenflash-health`, `greenflash-verify`, …) read it from the sibling `greenflash` folder.
This means the feature skills depend on the `greenflash` entry skill being installed alongside them — which the documented installs (`npx skills add greenflash-ai/agent-skills` and the plugin) always do, since both install the whole pack. When editing the config, keep it backward-compatible with already-installed skill versions.
## Documentation
Full API and plugin documentation: https://docs.greenflash.ai/features/claude-code-skill