Install
$ agentstack add skill-growthbook-skills-gb-setup ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ● Environment & secrets Used
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
gb-setup
One-skill onboarding for the GrowthBook plugin. Walks the user through GB_API_KEY and an optional GB_API_URL for self-hosted. Validates the credentials by hitting the live API, then writes ~/.config/growthbook/.env with chmod 600.
The API key is a Personal Access Token (PAT) tied to a GrowthBook user, so the API attributes any flags or experiments the write skills create to that user automatically — there's no separate owner identifier to configure.
gb-call reads this file when the corresponding environment variables aren't set, so the user gets a one-time config rather than editing their shell rc. Real environment variables always win over the file — useful for CI and one-off overrides.
Workflow
- Detect current state. Check what's already configured. Don't ask the user for values they already have unless they want to change them.
``bash test -f ~/.config/growthbook/.env && echo "exists" || echo "missing" ``
If the file exists, read it with the Read tool, parse the KEY=value lines, and surface a masked summary:
``` I see existing config at ~/.config/growthbook/.env: GBAPIKEY = gbpat****wxyz (last 4 shown) GBAPIURL = (not set — defaults to https://api.growthbook.io)
Want to keep these, update one, or start fresh? ```
Also note what's in process.env — if the user has shell exports, those will override the file. Surface that ("GB_API_KEY is also set in your shell environment; the file value won't be used unless you unset the shell var.")
- Collect
GB_API_KEY(required). If keeping the existing value, skip. Otherwise, show the transcript-exposure notice first so the user can make an informed choice:
> Before you paste your key: anything you type into this chat is stored in your local Claude Code transcript (~/.claude/projects/...) and sent to Anthropic as part of the conversation. The skill will mask the key in its replies, but the value you paste cannot be masked retroactively. > > Recommendation: generate a fresh PAT for this plugin rather than reusing your personal admin token. That way you can revoke it independently if anything goes wrong, without affecting your other API access.
Then ask:
> Paste a Personal Access Token (PAT) or Secret Key. Get one at: > - Cloud: > - Self-hosted: https:///account/personal-access-tokens > > PATs start with gb_pat_; Secret Keys with secret_. Either works.
Once captured, never echo the value back in any later step — mask all but the last 4 characters.
- Ask about self-hosted (optional
GB_API_URL).
> Are you using GrowthBook Cloud (api.growthbook.io) or a self-hosted instance?
- Cloud → skip, leave
GB_API_URLunset (gb-call defaults tohttps://api.growthbook.io). - Self-hosted → prompt for the API base URL (e.g.
https://api.acme-internal.com).
Validate the shape before accepting:
- Must parse as a URL (i.e.
new URL(value)does not throw). - Scheme must be
https://(refusehttp://— auth headers over cleartext leak credentials; refusefile://,data://, etc.). - Hostname must be non-empty.
- No path component —
new URL(value).pathnamemust be/or empty. A value likehttps://api.acme.com/v1would silently mis-route every request; reject with a message that asks for just the host. - Strip a single trailing slash before storing.
Refuse any URL that fails these checks and loop back to ask again; don't silently coerce. Silent fix-ups train users to trust that the system "just works" when the value is sometimes wrong in ways the system can't fix.
- Validate against the live API. Set the collected values as ad-hoc env vars and call a lightweight authenticated endpoint:
``bash GB_API_KEY='' \ GB_API_URL='' \ ${CLAUDE_PLUGIN_ROOT}/scripts/gb-call GET /api/v1/projects ``
Interpret the result:
- 2xx → credentials work. Move on.
- 401 / 403 → "Authentication failed. The API key is invalid, expired, or has been revoked. Generate a fresh one at `` and try again." Loop back to step 2.
- 404 on a
growthbook.iohost → "Got a 404 fromapi.growthbook.io. If you're actually on self-hosted, setGB_API_URLto your API base URL." Loop back to step 3. - Network error / DNS → "Couldn't reach ``. Check the URL and your network." Surface the raw error.
Don't proceed past validation. If the user is stuck, halt and let them debug — don't write a broken config.
- Write
~/.config/growthbook/.env— order matters for security.
Create the directory and lock it down first:
``bash mkdir -p ~/.config/growthbook chmod 700 ~/.config/growthbook ``
The directory's 0700 (owner-only) permission is what protects the file from being readable by other users on the system during the next step. Writing the file first and then chmod-ing it leaves a window where the file inherits the user's umask (typically 0022 → mode 0644, world-readable). Locking the directory closes that window — files inside a 0700 directory aren't reachable by other users regardless of file mode.
Then write with the Write tool. Format — one KEY=value per line, no quoting (our values never contain spaces, =, or newlines):
`` GB_API_KEY= GB_API_URL= ``
Then set the file mode as belt-and-suspenders:
``bash chmod 600 ~/.config/growthbook/.env ``
0600 means owner-read/write only. With the directory at 0700 and the file at 0600, the secret is unreachable to any other user on the system.
- Report. Tell the user:
- Where the file is (
~/.config/growthbook/.env). - What's in it (mask the API key — show only last 4).
- That env vars take precedence over the file (so CI / one-off overrides keep working).
- That re-running
/growthbook:setupupdates the file. - The next thing they probably want:
/growthbook:flag-searchto see their flags, or/growthbook:experiment-brainstormto look at past results.
Guardrails
- Never echo the API key back in plain text. Always mask except for the last 4 characters. The skill's output ends up in the user's terminal and transcript.
- Surface the transcript-exposure risk before the user pastes a key. The value typed into the chat is stored in
~/.claude/projects/...and sent to Anthropic as part of the conversation; the skill cannot retroactively redact it. Always recommend a freshly-scoped PAT over reusing an admin token. This isn't paranoia — it's the right way to handle a workflow that requires a user to paste a secret into a conversational interface. - Revocation guidance is a real fix, not a footnote. If a key was ever exposed, the only effective remediation is to revoke and rotate it at
/account/personal-access-tokens. Surface this whenever the user expresses concern about an old or shared key. - Directory at
0700before file write; file at0600after. The order matters. The Write tool inherits the user's umask, which on most systems creates files at mode0644(world-readable). Locking the directory down first means even the brief window between Write and chmod is not reachable by other users. Then chmod the file as defense in depth. Both steps are non-optional; skipping either turns the PAT into a leak. - Env vars take precedence over the file. If
process.env.GB_API_KEYis set, gb-call ignores the file value. Surface this if both are configured so the user understands what's actually in effect. - Validate against a real endpoint, not just shape. A token that "looks right" but doesn't work is worse than no token — the user thinks they're set up.
GET /api/v1/projectsis lightweight, requires auth, and works on every plan tier. - No owner identifier to collect. The PAT is tied to a GrowthBook user, so the write skills (
flag-create,experiment-launch) let the API attribute new flags and experiments to the token's user automatically. Don't prompt for an email or userId during setup. - Don't strip leading whitespace from line values, but strip trailing. A PAT shouldn't have whitespace at all; if the user pasted one with a trailing newline from a clipboard, strip it. Leading whitespace is unusual enough that surfacing rather than silently fixing is safer.
- Don't write a partial file. If validation fails, halt without touching
~/.config/growthbook/.env. The user re-runs setup; we never overwrite a working config with a broken one. - Strip trailing slash on
GB_API_URL.gb-calldoes this too, but doing it here avoids "the file has a slash and the URL has a slash" double-slash bugs in logs. - Don't propose env vars we don't actually use. Only
GB_API_KEYandGB_API_URL. Anything else is scope creep and will confuse future contributors.
File format reference
~/.config/growthbook/.env:
GB_API_KEY=gb_pat_abc123def456
GB_API_URL=https://api.your-self-hosted.com
Lines starting with # are comments. Blank lines are ignored. No quoting; values are taken verbatim from the first = to end of line.
Endpoints used
GET /api/v1/projects— validation only. Lightweight, requires auth, available on every tier.
Handoffs
flag-search— natural first call after setup completes (no required inputs, exercises the new config).experiment-brainstorm— if the org already has stopped experiments, surfaces them immediately.- Any skill that emits a config-related error points back here.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: growthbook
- Source: growthbook/skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.