Secret Sentinel

List the active secret detectors and allowlist rules. Returns: {"count": int, "patterns": [{"name", "severity", "advice"}], "entropy_detector": {...threshold and advice...}, "allowlist_rules": [{"name", "description"}]} Raw regexes are intentionally not exposed.

Return the text with every detected secret replaced by [REDACTED:<detector>]. Use it before quoting logs, stack traces, configs, command output or diffs that may contain credentials, in the conversation, an issue, a commit message or a document. Everything that is not a secret is kept exactly as it was. Args: text: The text to clean. Returns: {"text": the redacted text, "redactions": how many secrets were replaced, "by_pattern": {detector name: count}}

Audit AI agent and MCP client configuration files for hardcoded secrets. Checks the project's .mcp.json, .claude/settings*.json, .cursor/mcp.json, .vscode/mcp.json, .gemini/, .codex/ and other client configs, plus agent instruction files (CLAUDE.md, AGENTS.md, GEMINI.md, .cursorrules...), and the user-level configs of Claude Code, Claude Desktop, Cursor, VS Code, Cline, Windsurf, Gemini CLI, Codex, Kiro, Continue, Zed and others. Project files report whether git tracks them: a key in a tracked .mcp.json ships with the repository. Args: project_path: Absolute path to the project to audit (optional). include_user: Also scan configs in the user's home folder (default true). include_transcripts: Also scan the newest Claude Code / Codex session transcripts (default false; slower). max_findings: List at most this many findings, most severe first. Returns: The standard redacted report plus "agent_files": [{"client", "scope", "path", "findings", "tracked_by_git", "hint"}] for every file found.

Recursively scan a directory tree for exposed secrets. Inside a git repository the file list comes from git (tracked files plus untracked files no .gitignore excludes), so every ignore rule is honoured. Always skips .git, node_modules, virtualenvs, caches, dist, build, minified bundles, lockfiles, binaries, files over 5 MB and the paths the project's .secret-sentinel.toml excludes. Args: path: Absolute path to the directory to scan. max_files: Stop after scanning this many files (default 500). max_findings: List at most this many findings, most severe first (default 200, 0 = no limit). Returns: The standard redacted findings report; finding paths are relative to the scanned root, using forward slashes.

Scan a single file for exposed secrets. UTF-8, UTF-16 and UTF-32 text is decoded by byte-order mark. Binary files (null-byte heuristic) and files larger than 5 MB are skipped and reported as such in the summary. The project's .secret-sentinel.toml and baseline apply. Args: path: Absolute path to the file to scan. max_findings: List at most this many findings, most severe first (default 200, 0 = no limit). Returns: The standard redacted findings report (see scan_text). Returns an error result if the file does not exist.

Scan the lines added by the most recent commits (git log -p). Each finding is tagged with the short hash of the commit that introduced it. A secret that was later deleted is still reported — history retains it, so the credential must be rotated regardless. Args: repo_path: Absolute path to a git repository (or any path inside one). max_commits: How many commits back to inspect (default 50). all_branches: Walk every branch, tag and the stash instead of only the history of HEAD (default false). max_findings: List at most this many findings, most severe first (default 200, 0 = no limit). Returns: The standard redacted findings report, with a "commit" field on each finding.

Scan the commits in base..head: by default, what `git push` would send. Run this before pushing. With the defaults it scans every commit on the current branch that its upstream does not have yet. If the branch has no upstream, the error says so: pass the branch you will push to as base (for example "origin/main"). Args: repo_path: Absolute path to a git repository (or any path inside one). base: Commits reachable from here are excluded (default "@{upstream}"). head: Last commit to include (default "HEAD"). max_commits: Scan at most this many of the newest commits in the range. max_findings: List at most this many findings, most severe first (default 200, 0 = no limit). Returns: The standard redacted findings report with a "commit" field on each finding, plus "commits_scanned" and "range".

Scan ONLY the lines currently staged for commit (git diff --cached). This is the pre-commit checkpoint: it inspects exactly the content the next commit would publish, and reports the file and post-commit line number of every added secret. Local diff settings (external diff tools, textconv filters, prefixes, path quoting) are overridden, and git runs no configured programs. Args: repo_path: Absolute path to a git repository (or any path inside one). max_findings: List at most this many findings, most severe first (default 200, 0 = no limit). Returns: The standard redacted findings report. If nothing is staged the result is clean with an explanatory summary.

Scan a snippet of text or code for exposed secrets. 64 detectors: GitHub/GitLab/npm/PyPI tokens; OpenAI, Anthropic, xAI, DeepSeek, Perplexity, OpenRouter, Groq, NVIDIA, Hugging Face, Replicate, Together, Cerebras, LangSmith, Langfuse, Pinecone, Firecrawl and Tavily keys; AWS (incl. Bedrock), Azure, Google, DigitalOcean, Cloudflare, Vercel, Supabase, Neon, PlanetScale, Fly.io, Databricks, Vault and Doppler credentials; Stripe, Square, Shopify, SendGrid, Resend, Mailchimp, Twilio, Slack, Discord, Telegram, Linear, Notion, Atlassian, Postman, Sentry, Grafana and PostHog tokens; JWTs, private and age keys, credentialed connection strings and URLs, generic password/secret/token assignments, plus high-entropy strings. Placeholders ($VAR, ${VAR}, {{ templates }}, changeme, masked values...) are skipped, and a line carrying "secret-sentinel: ignore", "pragma: allowlist secret" or "gitleaks:allow" is counted in "suppressed" instead of reported. Args: text: The raw text to scan (code, config, diff output, logs...). source_name: Label used in each finding's "file" field. max_findings: List at most this many findings, most severe first (default 200, 0 = no limit). When capped, the result adds truncated, total_findings, counts_by_severity, counts_by_pattern and top_files. Returns: {"clean", "findings": [{"file", "line", "pattern", "severity", "redacted", "advice"}], "files_scanned", "summary"}. Secret values are ALWAYS redacted (at most 4 chars, never more than a quarter of the value, + length); the full value is never included.