tailtest — the test + security validator that lives inside Claude Code. Never blocks, never lies.
Project description
tailtest
The test + security validator that lives inside Claude Code. Never blocks, never lies.
tailtest watches every edit your AI agent makes inside Claude Code, runs the tests that matter, scans for security issues that matter, and feeds findings back into Claude's next turn so the agent can fix them in the same session. The hot loop never blocks your work.
Current release: v0.4.1 — hook format fix. Corrects SessionStart output (plain text, not JSON) and PostToolUse envelope (additionalContext at top level, not wrapped in hookSpecificOutput). Fixes PostToolUse crash on ImportError; tightens bootstrap import guard. Hooks now work correctly in interactive sessions.
Quickstart (5 minutes)
# 1. Install the plugin from the GitHub marketplace (no pip needed)
claude plugin marketplace add avansaber/tailtest
claude plugin install tailtest@avansaber-tailtest
# 2. Restart your Claude Code session
# (the skill registry doesn't hot-load, so a restart is mandatory)
# 3. Open a project and let Claude edit a Python or JS/TS file
# tailtest's hot loop fires automatically on every Edit/Write
After the first edit, look for tailtest: N/N tests passed · M.Ms in Claude's next-turn context. That is the hot loop talking to Claude. The full output lands at .tailtest/reports/latest.html and .tailtest/reports/latest.json.
Want the standalone CLI too? pip install tailtester gives you tailtest run, tailtest doctor, and tailtest scan outside of Claude Code. It is optional for plugin users.
For the full walkthrough including troubleshooting, see docs/quickstart.md. For install gotchas (PEP 668, v1 upgrade, hook Python resolution), see docs/install.md. For the full config schema, see docs/configuration.md.
What tailtest does today
- Runs your tests on every edit, using the project's native runner (pytest for Python; vitest or jest for JS/TS) with native test impact analysis so only the affected tests run.
- Scans for secrets via gitleaks on every changed file. CWE-798 (hardcoded credentials) tagged automatically.
- Scans for SAST issues via Semgrep across the changed files using the curated
p/defaultruleset. Configurable per project. - Scans dependencies for known vulnerabilities via the OSV.dev API on every manifest edit (
pyproject.toml,package.json). Hydrated severity, CWE IDs, fixed-version hints, and alias dedup so you see each advisory once. - Computes delta coverage on the lines your edit touched (Python). The next-turn context calls out exactly which new lines are uncovered.
- Generates context-aware starter tests via
/tailtest:gen <file>. AST-scans the source for domain entities (enums, exception types, key classes, auth patterns), loads the project profile for runner and framework context, and generates tests that understand your code's domain. Pass--context "description"to describe the module manually. Line 2 of the output is a detection note showing what context was used. - Filters known issues via a baseline at
.tailtest/baseline.yaml. Existing debt stays silent; only NEW issues surface. - Renders a self-contained HTML report at
.tailtest/reports/latest.htmlafter every run. No JavaScript, no CDN, opens offline. - Red-teams AI agent entry points at
paranoiddepth via a 64-attack catalog covering the OWASP LLM Top 10. Each attack is judged byclaude -p; findings land in a timestamped HTML report at.tailtest/reports/redteam-*.html. - Never blocks your work. tailtest reports; you decide.
Slash commands
After install, Claude Code knows the following user-invocable skills:
| Command | What it does |
|---|---|
/tailtest:status |
Compact one-screen status: depth, runner, last-run summary, delta coverage, next action |
/tailtest:report |
Full detail of the last run; opens the path to latest.html so you can read it in a browser |
/tailtest:security |
Current security posture: which scanners are on, ruleset, new vs baselined finding counts |
/tailtest:debt |
Review the accepted-debt baseline (.tailtest/baseline.yaml) |
/tailtest:scan |
Re-scan the project profile from scratch |
/tailtest:gen |
Generate a starter test for an uncovered function |
/tailtest:depth |
Change the hot loop depth (off/quick/standard/thorough/paranoid) |
/tailtest:setup |
Onboarding interview that writes .tailtest/config.yaml |
/tailtest:memory |
View or clear the validator memory file (.tailtest/memory/validator.md) |
/tailtest:help |
Lists all skills with descriptions and links to docs |
/tailtest |
Show active project recommendations; dismiss or accept with an ID |
Configuration
Optional. tailtest works out of the box with sensible defaults. To customize, drop a .tailtest/config.yaml in your project root:
schema_version: 1
depth: standard # off | quick | standard | thorough | paranoid
security:
secrets: true # gitleaks
sast:
enabled: true
ruleset: p/default # any Semgrep ruleset id, including p/owasp-top-ten or p/ci
sca:
enabled: true
use_epss: false # opt-in to EPSS scoring (off until EPSS.io integration ships)
block_on_verified_secret: false
notifications:
auto_offer_generation: true # offer /tailtest:gen suggestions in the hot loop
Phase 1 configs (with sast: true/false as plain bools) keep parsing — the loader migrates them to the nested form transparently.
Distribution channels
| Channel | Install | When to use it |
|---|---|---|
| Claude Code plugin (recommended) | claude plugin marketplace add avansaber/tailtest && claude plugin install tailtest@avansaber-tailtest |
Full experience: hot loop hooks, skills, MCP, on-disk output. No pip required. |
| Standalone CLI (optional) | pip install tailtester |
CI pipelines, raw terminal use, or any workflow outside Claude Code. The PyPI package name is tailtester; the importable Python package is tailtest. |
| MCP server | tailtest mcp-serve |
Cursor, Windsurf, Codex, or any MCP-aware IDE. Requires the pip install. Phase 4 will harden this path. |
Install notes
- pip is optional for plugin users. The plugin is self-contained; the bootstrap finds the tailtest engine inside the plugin tree. You only need
pip install tailtesterif you want the standalone CLI or MCP server outside Claude Code. - Restart Claude Code after installing or upgrading the plugin. The skill registry and hook registry are frozen at session start; new plugins aren't picked up mid-session.
- macOS + Homebrew Python users: if you do install the standalone CLI, PEP 668 means
pip install tailtesterneeds--break-system-packagesor (better)pipx install tailtesterto keep tailtest in its own venv. - Upgrading from v1 (
tailtester0.2.x)? Uninstall the v1 package first (pip uninstall tailtester). The v1 and v2 packages share thetailtest.hookimport path; without the uninstall step, v1 shadows v2.
What tailtest does NOT do (yet)
Current limitations in v0.4.1:
- No multi-language SCA beyond Python + JS. Go / Rust / Java on the roadmap.
- No EPSS / KEV / NVD severity enrichment for SCA findings.
- No SCA discovery for projects without a
pyproject.tomlorpackage.json.
Repository layout
tailtest/
├── .claude-plugin/plugin.json Claude Code plugin manifest
├── .mcp.json MCP server wiring
├── hooks/ PostToolUse / SessionStart hook shims
├── skills/ User-invocable slash commands
├── src/tailtest/ Python package source
├── tests/ Pytest suite (1169 tests)
├── pyproject.toml
├── README.md This file
├── LICENSE Apache 2.0
├── CHANGELOG.md
└── SECURITY.md
Contributing
The project is in active development through Phase 8. The current release is v0.4.0. The issue tracker on github.com/avansaber/tailtest is open for bug reports and feature requests. PRs are reviewed but the bar for accepting external code is high while the architecture is still moving.
License
Apache 2.0. Copyright 2026 AvanSaber Inc.
Security
See SECURITY.md for how to report a vulnerability. tailtest practices what it preaches: every release passes a hygiene audit (gitleaks, trufflehog, manual review) before it ships, and the security layer dogfoods itself by scanning the public source tree on every commit.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file tailtester-0.4.1.tar.gz.
File metadata
- Download URL: tailtester-0.4.1.tar.gz
- Upload date:
- Size: 362.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: uv/0.8.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
fe7ea227a0ccdb05c8ee24f53a46f203fbfa10ef24f3c3d1eb5ba95517268d65
|
|
| MD5 |
f25f4f632be20aef7fbf346e256ebcd1
|
|
| BLAKE2b-256 |
4eb74e426d4d9c5f81117defa54018e7d416bc969966339eca103678096d212f
|
File details
Details for the file tailtester-0.4.1-py3-none-any.whl.
File metadata
- Download URL: tailtester-0.4.1-py3-none-any.whl
- Upload date:
- Size: 265.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: uv/0.8.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
5cda611a4df054fa1cdf721c8cb29b1391638632a7f2672e41ac7456780de27d
|
|
| MD5 |
e5fe99d1b83cd7bca3adfd8cd13b3400
|
|
| BLAKE2b-256 |
c6998268eea309643dfe7b5e1134b8d6110d1b31f946c0a1f61259bcc6b9260d
|