Slash commands
Type / in the message box to see WeHarness’s commands, grouped by area. A command is a shortcut for a request you could also write in your own words, in any language. Arguments offer suggestions: the session’s files, its branches, or catalog entries.
A command that needs something the session does not have yet (a repository, a GitHub repository, an approved environment, changes in the draft, a pull request) says so instead of running.
Review
Section titled “Review”| Command | What it does | Argument | Needs |
|---|---|---|---|
/review |
Reviews the project’s agent layer with WeHarness’s guidance. | A folder, file or concern (optional) | A repository |
/explain |
Explains a file or folder and how it fits the project. | A file or folder (required) | A repository |
| Command | What it does | Argument | Needs |
|---|---|---|---|
/plan |
Plans what to build: requirements, acceptance tests, options. See Plans and approval. | What to build or change (optional) | — |
/catalog |
Finds what the WeHarness catalog has for a need, and what fits here. | What you need, for example retries, approvals, tracing (required) | — |
Integrate
Section titled “Integrate”| Command | What it does | Argument | Needs |
|---|---|---|---|
/test |
Runs the project’s tests in the session’s sandbox. | Which tests or what to check (optional) | A repository |
/env |
Shows or proposes how clean checks set up and test this repository. | — | A repository |
/verify |
Checks the changes in a fresh sandbox with the approved environment (a clean check). | — | A repository, an approved environment and changes |
/push |
Asks to push the changes as a pull request. You approve it. | A pull request title (optional) | A GitHub repository and changes |
/pr |
Shows the session’s pull request and what changed since the push; prepares the next push when there are changes. You approve it. | — | A GitHub repository |
/branch |
Switches the session to another branch or tag. | A branch or tag (required) | A GitHub repository |
/checks |
Reads the pull request’s CI checks and what failed. | — | A pull request |
/use |
Brings one catalog entry into the work and checks the code against it. | A catalog entry (required) | A repository |
Ecosystem
Section titled “Ecosystem”Each of these reviews one area of the agent layer against the WeHarness catalog. It recommends what to keep, add or change, with file names and the catalog entries it relies on, and does not change code. Each takes an optional focus.
| Command | Area |
|---|---|
/models |
Model providers and APIs, adapters, fallbacks and routing (web app only) |
/frameworks |
Agent frameworks and SDKs the application builds on |
/harnesses |
Agent harnesses, coding harnesses and their improvement |
/orchestration |
Workflows, durable execution, waits, handoffs and agent-to-agent protocols |
/tools |
Tool contracts, MCP servers and clients, sandboxes |
/memory |
Agent memory, retrieval and context between steps |
/evaluations |
Evaluations, tracing and telemetry of agent behaviour |
/skills |
Procedural skill bundles that guide coding agents |
Evaluate
Section titled “Evaluate”| Command | What it does | Argument | Needs |
|---|---|---|---|
/remember |
Keeps something for your later sessions: a preference, a decision, a fact. See Memory. | What to remember (required) | — |
/traces |
Looks at what your running app did: recent traces and failures. See Tracing. | What to look into (optional) | — |
/eval |
Runs your agent’s scenario cases on the draft and compares with the last run. See Evaluations. | A behaviour to check (optional) | A repository |
/learn |
Keeps a correction as a lesson, with a case that stops it coming back. See Lessons. | What went wrong and what is right (optional) | A repository |
Commands in coding tools
Section titled “Commands in coding tools”The same commands, except /env, /verify, /push, /pr, /branch, /checks, /remember, /traces, /eval, /learn and /models, are offered to coding tools as MCP prompts. Coding tools also get a pr prompt of their own: the pull request of the WeHarness session they work on, without pushing. Tools that list prompts show them in their own menu. See Tools and prompts reference.
A command never approves anything. Plans, environments, pushes and lessons are approved by a person in the web app, whatever command started them.