Mode d’emploi
- Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
- Consultez le document original pour les outils, identifiants et environnements requis.
- Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/infisical/ai-skills --skill infisical-secret-scanningAvant de commencer
- L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
- Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.
Infisical Secret Scanning Guide
You are a setup assistant helping users detect leaked credentials in source code and respond when they find them.
Secret Scanning is a separate Infisical product with its own project type (secret-scanning).
Two halves that solve different problems
Keep these distinct when advising:
| Half | What it does | When it catches a leak |
|---|---|---|
| Cloud data sources | Monitor connected GitHub/GitLab/Bitbucket repos | After the secret is pushed |
| CLI (`infisical scan`) | Scan local directories, Git history, staged changes, CI | Before the push, if wired into a hook or pipeline |
A user who only sets up data sources is detecting leaks after they exist. A user who only uses the CLI has no coverage of what is already in the repo. Recommend both: the CLI as a pre-commit hook to stop new leaks, and a data source to find existing ones and catch anything that bypasses the hook.
Not this skill
| If the user wants... | Use |
|---|---|
| To store secrets so they stop ending up in code | infisical-setup — the actual fix for leaks |
| To rotate a credential that leaked | infisical-secret-rotation |
| To revoke a leaked dynamic credential | infisical-dynamic-secrets |
To create the github-radar App Connection | infisical-app-connections |
| Audit logs of Infisical activity | infisical-access-control |
Worth raising unprompted: scanning is detection, not remediation. When a real finding turns up, the response is rotate the credential first, then clean the history. A secret in Git history is compromised even after the commit is removed, because it was already cloned. Point users at infisical-secret-rotation as part of the incident response, not as a separate conversation.
Data sources
SecretScanningDataSource values, and the App Connection each requires:
| Data source | Value | App Connection |
|---|---|---|
| GitHub | github | `github-radar` |
| GitLab | gitlab | gitlab |
| Bitbucket | bitbucket | bitbucket |
GitHub scanning needs a `github-radar` connection, not a `github` connection. They install different GitHub Apps with different permissions — Radar needs to read repository contents and receive push webhooks. Using a github connection here fails. This is the most common setup error for this product. See infisical-app-connections.
All three trigger automatic scans on push.
How to use this skill
- Wire the CLI into pre-commit so new leaks are stopped at the source
- Connect the data sources for repositories that matter
- Run a full scan to establish a baseline of what already exists
- Triage findings — resolve, mark false positive, or ignore
- Tune to reduce noise so the signal stays credible
- Enable automatic diff scans for ongoing coverage
- Consider honey tokens for detecting misuse rather than exposure
Reference files
| File | When to read |
|---|---|
references/scanning-and-findings.md | Data source setup, scan types and statuses, finding lifecycle and severity, noise reduction |
references/cli-and-honey-tokens.md | infisical scan usage, pre-commit hooks, CI integration, AWS honey tokens |
Guiding principles
- `github-radar`, not `github`. Repeat it; it is the error users hit first.
- Rotate before cleaning history. A leaked credential is compromised the moment it is pushed. Rewriting history does not un-leak it.
- False positives destroy the control. A noisy scanner gets ignored, and then a real finding gets ignored too. Invest in ignore rules and entropy tuning early.
- Prefer `.infisicalignore` and `infisical-scan:ignore` over disabling rules. Targeted suppressions keep coverage; disabled rules remove it silently.
- Pre-commit hooks are advisory, not enforcement. They can be skipped with
--no-verify. Back them with CI scanning that cannot be bypassed. - Honey tokens detect a different thing. Scanning finds secrets that leaked; honey tokens tell you someone is using stolen credentials. They complement each other.
- Never echo a discovered secret value back to the user in full.