# How to Work With Mogplex (/guides/how-to-work-with-mogplex)



Good Mogplex requests are operationally specific. They name the target, the
desired outcome, the limits, and the proof you expect back.

This matters because the same request can travel through different surfaces:
the CLI composer, a GitHub mention, a Trigger, an Assignment, an Automation, or
a headless run.

## Give the agent a real job [#give-the-agent-a-real-job]

Start with the outcome, not the implementation detail:

```text
Find why the checkout flow fails after billing redirect. Reproduce the failing
path first, then patch the smallest code path that owns the redirect.
```

That is stronger than:

```text
Fix checkout.
```

The useful version tells Mogplex what to inspect, what order to follow, and what
kind of fix is acceptable.

## Include the operating boundaries [#include-the-operating-boundaries]

Add boundaries when they matter:

* whether code edits are allowed
* whether the agent should only diagnose
* whether tests, lint, typecheck, or build should run
* whether the agent should open a PR
* whether the route should stay on the current branch or create a new one
* whether a GitHub reply should be posted or only local changes should be made

Examples:

```text
Diagnose only. Do not edit code. Find why the trigger did not start for this
repo and report the exact missing state.
```

```text
Fix the failing PR check on this branch. Run lint and typecheck before pushing.
Do not change unrelated files.
```

## Name the source of truth [#name-the-source-of-truth]

Mogplex works best when the request says where truth lives.

Good sources of truth include:

* a GitHub PR, issue, or review thread
* a failing CI job
* a local command output
* a route in the web app
* a CLI panel or slash command
* an API endpoint
* a design or screenshot

Example:

```text
Use the latest unresolved PR review comments as scope. Fix only the true
correctness issues and leave style-only suggestions alone.
```

## Use exact mention syntax [#use-exact-mention-syntax]

GitHub mention routing is exact:

```text
@mogplex
@mogplex/triage
@mogplex/frontend-review
```

Use a bare `@mogplex` when the default mention route should decide what to do.
Use `@mogplex/<agent-slug>` when you want a specific agent route.

This is not the same as:

```text
@mogplex triage this
```

That is still a bare mention with text after it, not a targeted route.

Read next: [GitHub Mention Routing](/web/guides/github-mention-routing).

## Choose request shape by surface [#choose-request-shape-by-surface]

### CLI [#cli]

Use the CLI for repo-local work where you want to supervise approvals, diffs,
model choice, cost, and timeline events.

```text
/run reproduce the failing activation flow, identify the owning route, patch it,
then run the focused test.
```

Use slash commands when the task is about cockpit state:

* `/model` for model choice
* `/permissions` for permission mode
* `/diff` for patch inspection
* `/cost` for token and cost detail
* `/export` for a portable run artifact

Read next: [Slash Commands](/cli/guides/slash-commands).

### Trigger [#trigger]

Use a short request when the event already contains most of the context.

```text
@mogplex/ci-fixer investigate this failure and propose the smallest fix.
```

If the agent should mutate code, say that explicitly. If it should only report,
say that explicitly.

### Assignment [#assignment]

Assignments should encode stable repo responsibility, so prompts should describe
the durable operating rule.

```text
For this repo, review opened PRs for correctness regressions, missing tests,
and security-sensitive changes. Do not comment on formatting-only issues.
```

When the work changes often, prefer a one-off Trigger or Automation instead of
overloading the assignment.

### Automation [#automation]

Automations should separate decisions from mutations.

A good workflow request names the entry event, the role of each agent, and the
condition for moving to the next step:

```text
On PR opened, review for correctness issues. If real issues are found, branch to
an edit agent that patches them and opens a follow-up PR. Otherwise end after
the review summary.
```

Read next: [Automation Agent Roles](/web/automations/agent-roles).

### Headless run [#headless-run]

For `mogplex run`, put the whole job in the prompt and make the terminal
success condition clear:

```bash
mogplex run --repo "$REPO_ID" --create-branch \
  "Run the focused failing test, patch the root cause, rerun the focused test,
  and summarize the changed files."
```

Read next: [Headless Runs](/cli/automation/headless-runs).

## Ask for evidence [#ask-for-evidence]

A good request says what proof should come back:

* command output
* screenshot
* test name and result
* route or file path inspected
* PR number
* run ID
* API status code
* observed model or token state

Example:

```text
After the fix, tell me the exact command you ran and whether it passed.
```

## Keep reusable knowledge out of one-off prompts [#keep-reusable-knowledge-out-of-one-off-prompts]

If an instruction should apply repeatedly, put it in the right reusable layer:

* [Agents](/web/agents) for durable agent identity, prompt, slug, and model
* [Repo Instructions](/configure-and-extend/repo-instructions) for local repo
  conventions, commands, and verification rules
* [Custom Slash Commands](/configure-and-extend/custom-slash-commands) for
  reusable CLI prompt templates
* [Skills, Rules, and Context](/web/agents/skills-rules-context) for shared
  procedures, policy, and background knowledge
* [Assignments](/web/assignments) for repo-owned standing work
* [Automations](/web/automations) for repeatable workflow graphs

One-off prompts are best for the current run. Durable behavior belongs in the
product surface that owns it.

## Useful request templates [#useful-request-templates]

### Diagnose only [#diagnose-only]

```text
Diagnose only. Do not edit code. Reproduce the reported behavior, identify the
owning code path or configuration state, and brief me with confirmed findings.
```

### Patch and verify [#patch-and-verify]

```text
Patch the smallest code path that owns this behavior. Run the focused check
first, then the relevant lint/type/build check. Summarize changed files and
validation.
```

### Review [#review]

```text
Review this PR for correctness bugs, regressions, missing tests, and risky
edge cases. Put findings first with file and line references. Keep style-only
comments out unless they hide a real defect.
```

### Route a GitHub mention [#route-a-github-mention]

```text
@mogplex/triage classify this issue, identify the likely owner, and suggest the
next concrete step. Do not edit code.
```

### Create a repeatable workflow [#create-a-repeatable-workflow]

```text
Build an automation for PR-opened events: review first, branch to an edit agent
only when there are confirmed issues, then stop after a summary if no edit is
needed.
```

## Read next [#read-next]

<Cards>
  <Card title="Common Use Cases" href="/guides/common-use-cases" />

  <Card title="Choose the Right Routing Surface" href="/web/guides/choose-routing-surface" />

  <Card title="GitHub Mention Routing" href="/web/guides/github-mention-routing" />

  <Card title="Slash Commands" href="/cli/guides/slash-commands" />
</Cards>
