Thunderbolt is an open-source AI client that lets people connect their chosen models and run the client on their own infrastructure, including local or on-premises deployments. It is aimed at organizations and individuals who want to use different model providers while keeping control of their data.
Borrowing it
Nothing to install: this file belongs to thunderbird/thunderbolt. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/thunderbird/thunderbolt/main/.claude/skills/thunder-deep-review/SKILL.mdgit clone --depth 1 https://github.com/thunderbird/thunderboltWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/thunderbird/thunderbolt/thunder-deep-review)<a href="https://agentmods.dev/skills/thunderbird/thunderbolt/thunder-deep-review"><img src="https://agentmods.dev/badge/skills/thunderbird/thunderbolt/thunder-deep-review/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/thunderbird/thunderbolt/thunder-deep-review"><img src="https://agentmods.dev/badge/skills/thunderbird/thunderbolt/thunder-deep-review.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00000 | $0.07689 |
| Opus 5.5 | $0.00000 | $0.03076 |
| Sonnet 5.5 | $0.00000 | $0.01538 |
| Haiku 4.5 | $0.00000 | $0.00769 |
Grade A, and why
thunder-deep-review scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 21d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 185 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Thunder Deep Review
A read-only senior-grade PR reviewer for thunderbird/thunderbolt. It reproduces the house bar — an architecture-and-taste reviewer first, a linter second. It catches the high-value issues generic bug-bots miss: wrong abstraction layer, coupling, undocumented intent, convention drift, bloat/deps, and "this will haunt us" maintainability costs — alongside real correctness/security bugs.
The "house bar" is a standard, not a person. Never name or impersonate any individual. Encode the bar, not the human.
Deployment context — determine this FIRST
One engine, two products. Detect your mode from OBSERVABLE signals in the invoking prompt, then obey it everywhere the steps below split on mode:
- CI / gated mode — you are in this mode iff the invoking prompt does any of: supplies a deep-mode/bounded-mode flags file path, or requests structured JSON output against a provided schema and states that a downstream precision gate filters your candidates. Here you are the RECALL pass: a downstream precision gate (a second model step) filters your candidates and is allowed to drop most of them — so favor recall. Surface every grounded candidate even when unsure it clears the bar; report marginal or low-certainty findings with
confidence: lowinstead of dropping them. Your output is the structured JSON the invoking prompt defines — no markdown report, no severity trailer. - Local / ungated mode — anything else (a developer invoked you ad hoc). There is NO gate behind you: your rendered markdown report (exact format:
assets/finding-template.md) is the final artifact, so apply the full filtering pipeline below (Pass B, recall floors, noise suppression, self-validation).
When signals are ambiguous, assume LOCAL mode — an unfiltered candidate dump on a developer is worse than filtering twice.
In both modes: identifying the issue is the job; a suggested fix is optional, never required.
Operating rules (low freedom — always)
What ships with it
7 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 21d ago Changed · -3 lines 70a844131fad
- 23d ago Changed · +18 lines e9b38be241ae
- 2mo ago First seen · 170 lines · 0 tokens per session scan A fcdea08ad494
thunder-deep-review is a skill published in the GitHub repository thunderbird/thunderbolt (4,816 stars, last pushed today), licensed MPL-2.0. It costs nothing until one of its globs matches a file; then it loads 7,689 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
code-review
Review code for quality, maintainability, and correctness.
describe-pr
Only use when the user explicitly invokes this skill by name.
advisor-orchestrator-worker
Use when a task is too large for one model pass, needs parallel research or generation across many subtasks (like researching a dozen competitors at once), or the user asks to orchestrate multiple models, split work across a model team, run an advisor-worker loop, have a stronger model review the plan while cheap…
code-review
Systematic code review for bugs, security, style, and performance.
rag-techniques-contribution-guidelines
Checks a RAGTechniques pull request against the repository's contribution guidelines (CONTRIBUTING.md): where new technique notebooks go, the README list entry and table row every new technique needs, the required notebook structure, and notebook hygiene. Use on any PR that adds or changes a notebook under…
code-reviewer
Staff-engineer-level code review delivering 10 prioritized actionable findings across architecture, security, performance, and maintainability.