$ ~/registry/skill/facebook-fboss-skills-fboss-review

SKILL

fboss-review

Name: fboss-review
Author: facebook

Review FBOSS diffs with parallel checks for architecture, reliability, and testing.

Stars

★ 976

Source

GitHub

Updated

2026-06-27

// securityCaution

Prompt only, no code
Open source

Summary

This appears to be an open-source, prompt-only review skill with no required secrets, no declared remote endpoints, and no executable installation steps; overall risk is low. The main caveat is that the README instructs the host agent to inspect local diffs/reference files and possibly call other tools, but that is workflow guidance rather than direct capability of the skill itself.

CredentialsLow risk

The material explicitly states that no keys or environment variables are required, and it does not request API tokens, account credentials, or other long-lived secrets, so credential exposure and abuse risk appears low.

Network egressLow risk

No remote endpoint is declared, and the system checks mark it as prompt-only; while the README mentions fetching Phabricator diff details via another tool, this skill itself does not define a data egress target or direct networking logic.

Code executionLow risk

As a prompt-only skill, the material contains no scripts, binaries, installation commands, or local execution payloads. The README references `sl status`, `sl diff`, and tool calls as suggested workflow steps, which is not the same as built-in code execution capability.

Data accessLow risk

The skill is intended to review FBOSS diffs and references changed content plus some pattern files, but the material does not declare direct file I/O, bulk collection of user data, or access beyond what is normally needed for code review.

Supply chainCaution

The source is an auditable open-source GitHub repository, which materially lowers supply-chain risk; however, the license is unspecified, community adoption is shown as 0 stars, and maintenance status is unknown. Before use, it is still advisable to verify that the specific skill files actually map to the repository and to check recent maintenance activity.

Recommendations

Before onboarding, verify that the actual skill prompt/files genuinely come from the stated GitHub repository and review recent commits.
If the host agent may follow the README to call external tools or read local diffs, restrict accessible directories and allowed tools at the host level.
Ensure review results are not automatically posted back to code hosting or review systems; keep a human approval step before publishing.

Audited by: gpt-5.4 · 2026-06-16

// install

Copy the install command and let the AI configure it · recommended for beginners

Please install the "fboss-review" skill from askskill:
1. Download https://raw.githubusercontent.com/facebook/fboss/main/fboss/skills/fboss-review/SKILL.md
2. Save it as ~/.claude/skills/fboss-review/SKILL.md
3. Reload skills and tell me it's ready

// download

Download SKILL.mdInstall manifest ↗

// examples

Review a switch feature change

Prompt

Please review this FBOSS diff. Focus on reliability, silent failure risks, agent architecture impact, SAI/SDK compatibility, and whether test coverage is sufficient. List findings and fixes by severity, and do not auto-post comments.

Expected

A prioritized code review report with issues, impact analysis, and suggested fixes.

Check config and platform compatibility

Prompt

Analyze this FBOSS change related to configuration and platform behavior. Focus on platform/config, FSDB/thrift_cow interactions, cross-cutting design impact, and whether it introduces hard-to-observe failure modes.

Expected

A review checklist and improvement recommendations for platform compatibility and system design risks.

Assess test coverage gaps

Prompt

Review this FBOSS diff from engineering and test quality perspectives. Identify missing unit, integration, and regression test scenarios, and explain why each test is necessary.

Expected

A review focused on testing gaps, including recommended additional test cases.

// docs

FBOSS Review

Overview

Multi-reviewer code review for FBOSS diffs. Subsumes general review (/review-diff functionality) plus FBOSS-specific domain expertise.

Scope

Currently fboss/ only. TODO: extend to configerator/source/neteng/fboss, neteng/netcastle, neteng/fboss.

IMPORTANT: No Auto-Posting

All findings are shown to the user only. Never post to Phabricator automatically.

Review Workflow

Step 1: Identify Changes

Run sl status and sl diff to get changed files
If user specifies a diff number, fetch diff details via mcp__plugin_meta_mux__get_phabricator_diff_details

Step 2: Determine Applicable FBOSS Reviewers

Files in agent/, SwSwitch, HwSwitch -> Agent Reviewer (#6)
Files in fsdb/, thrift_cow/ -> FSDB/thrift_cow Reviewer (#7)
Files in platform/, config, sensor/fan -> Platform Reviewer (#8)
Files in sai/, hw/sai/, SDK -> SAI/SDK Reviewer (#9)
Test files or coverage-affecting changes -> Testing Reviewer (#10)
Changes spanning SwSwitch + HwSwitch, new SwitchState fields, new counters, BUCK target changes, new client infrastructure, or 3+ directory scope -> Architect Reviewer (#11)

Step 3: Dispatch Reviewers in Parallel

Use Agent tool (sonnet model) to dispatch all applicable reviewers simultaneously:

Generic reviewers 1-5: ALWAYS dispatch. Use personas from references/generic-reviewers.md.
FBOSS reviewers 6-10: Dispatch ONLY if their area is touched.
Each reviewer receives: full diff content + their persona instructions.
FBOSS reviewers additionally load patterns from ../fboss-code-standards/references/<area>-patterns.md.

#	Reviewer	Focus	Patterns File
1	Reliability	Error handling, RAII, logging, timeout/retry, graceful degradation	-
2	Engineering / Performance	Algorithmic complexity, unnecessary copies, lock contention, modern C++	-
3	Code Quality	Readability, modularity, duplication, API design	`../fboss-code-standards/references/general-patterns.md`
4	Summary & Test Plan	Title accuracy, summary completeness, test plan adequacy, diff coherence	-
5	Silent Failure Finder	Logic errors, lossy conversions, race conditions, silent data loss	-
6	Agent Architecture	Mono/multi-switch, state management, warmboot, HwSwitch/SwSwitch boundary	`../fboss-code-standards/references/agent-patterns.md`
7	FSDB & thrift_cow	State/Stats duality, COW node safety, subscriptions, build time	`../fboss-code-standards/references/thrift-cow-fsdb-patterns.md`
8	Platform & Config	Platform services, JSON configs, startup order, OSS sync	`../fboss-code-standards/references/platform-config-patterns.md`
9	SAI/SDK Integration	SAI API usage, object lifecycle, vendor SDK compat, attributes	`../fboss-code-standards/references/sai-sdk-patterns.md`
10	Testing Standards	Coverage, naming, fixtures, NSDB impact, HW test patterns	`../fboss-code-standards/references/testing-patterns.md`
11	FBOSS Architect	Cross-cutting design: layering, state derivability, abstraction, code org, infra reuse	`references/fboss-specific-reviewers.md`

Step 4: Verification

Dispatch a single verifier agent (loads references/verifier.md) to:

Deduplicate findings across reviewers
Calibrate confidence scores
Filter false positives
Threshold: only report findings with confidence >= 0.7

Step 5: Present Findings

Output a structured report:

| File:Line | Reviewer | Severity | Issue | Confidence |
|-----------|----------|----------|-------|------------|

Never post findings to Phabricator.

When to Load References

Topic	Action
Generic reviewer personas	Read `references/generic-reviewers.md`
Verification/dedup logic	Read `references/verifier.md`
FBOSS-specific reviewer personas	Read `references/fboss-specific-reviewers.md`

…

View full docs ↗

facebookGet→

$ loading_