Agent
Independent read-only architecture reviewer focused on long-term constraints beyond local code — module boundaries, duplication of capability, public-API/back-compat, migration expand/contract, rollout/rollback, and domain conventions. Use during parallel review of structural changes.
Agent
Independent read-only correctness reviewer. Use during parallel review of code changes to find broken invariants, bad state transitions, missing error propagation, races and edge cases introduced by the diff.
Agent
Independent read-only intent reviewer. Compares the diff against stated acceptance criteria and non-goals to find unmet requirements, violated non-goals, undeclared behavior, and missing required edge cases. Judges "is this what was asked", not "is this elegant". Use when an intent file or PR description exists.
Agent
Synthesizes verified verdicts, deterministic-gate state and risk tier into a single gate decision (PASS / PASSWITHNOTES / HUMANREQUIRED / BLOCK / REJECTINTAKE) plus a concise human report. Only verified findings may block. Use as the final step of the multi-review pipeline.
Agent
Cheap risk-tiering router for the multi-review pipeline. Reads REVIEW.md, scores the diff, assigns R0–R3, and selects the minimal sufficient sensor set. Use at the start of a code review to decide how much review fleet a change warrants.
Agent
Conditionally-triggered read-only runtime reviewer. Runs the narrowest safe targeted/repro commands to observe async timing, resource leaks, timeouts/retries, query counts and performance. Use for R2/behavioral changes where static reading is insufficient.
Agent
Independent read-only security reviewer organized around trust boundaries and data flow. Use during parallel review of changes touching auth, tenancy, injection surfaces, secrets/PII, dependencies, or user-controlled data reaching LLM prompts/tools.
Agent
Independent read-only test-integrity reviewer. Reads the test diff FIRST to catch deleted/skipped/loosened assertions, expectations rewritten to match wrong behavior, happy-path-only coverage, and implementation+test collusion. Use during parallel review of code changes.
Agent
Independently falsifies or verifies ONE candidate review finding from a clean context. Receives only the claim, location, target and reproduction hint — never the originating reviewer's reasoning. Use to confirm a candidate before it is allowed to block.
Agent
A specialist agent for diagnosing failures when code is compiled for a different operating system or processor, especially OHOS.
Agent
An OHOS work scheduler for OpenHarmony and related systems such as KaihongOS. It analyzes requests and assigns them to a suitable OHOS specialist agent.
Agent
A feasibility-analysis agent for moving Linux software to OpenHarmony, an operating system for connected devices. It assesses the expected difficulty and risks of the port.
Agent
An architecture-planning agent for moving software from Linux to OpenHarmony, an operating system for connected devices. It plans code changes and ways to adapt Linux-specific interfaces to OpenHarmony interfaces.
Agent
An agent role for operating OHOS source code and builds on remote servers over SSH. OHOS is an operating system project.
Agent
A runtime-error diagnosis agent for software running on an OpenHarmony device, an operating system for connected devices. It analyses why a program failed on the device.
Agent
An agent for exploring the source code and structure of Linux software before porting it to OHOS, an operating system platform.