Commands

The mentu-autopilot plugin registers five slash commands in Claude Code. Each command is designed for a specific stage of the bug-fix workflow.

/triage

Read-only dashboard. Fetches open bug memories from your Mentu workspace, scores them, and displays an actionable table. Does not create or modify any commitments.

Syntax

/triage

What It Does

Fetch — Pulls all open bug memories from the workspace.
5-Gate Garbage Filter — Each bug passes through five sequential gates. A failure at any gate eliminates the bug from the pipeline:
- Body Coherence — Does the report contain a meaningful description?
- Test Detection — Is there a reproduction path or test case?
- Project Match — Does the bug belong to this project’s domain?
- Duplicate Collapse — Is this a duplicate of an already-triaged bug?
- Actionability — Can this be fixed with the available codebase?
Score — Survivors are scored by severity, clarity, and estimated effort.
Display — Results are shown in a ranked table.

Example Output

Triage Results (12 fetched → 7 survived → 5 actionable)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

 #  Score  ID              Title                          Tier  Gate Failures
 1  9.2    mem_a8f3c21...  Login form rejects valid email  T1   —
 2  8.7    mem_b4d9e10...  Dashboard chart renders blank   T2   —
 3  7.4    mem_c1a2f33...  Export CSV missing headers       T1   —
 4  6.8    mem_d7e5b44...  Sidebar collapses on mobile     T2   —
 5  5.1    mem_e9c0a55...  Slow query on clients page      T3   —

Filtered (5):
  - mem_f1... "asdf" — failed Body Coherence
  - mem_f2... "test bug" — failed Actionability
  - mem_f3... duplicate of mem_a8f3c21
  - mem_f4... unrelated project domain
  - mem_f5... "colors look wrong" — failed Test Detection

No data is mutated. Run /fix <mem_id> on any survivor to begin the fix pipeline.

/fix

End-to-end single ticket fix. Takes a memory ID and drives it through the full pipeline: fetch, investigate, branch, fix, build, push, PR, and submit.

Syntax

/fix <mem_id>

Parameters

Parameter	Required	Description
`mem_id`	Yes	The Mentu memory ID (e.g., `mem_a8f3c21...`)

The 10 Phases

Fetch — Pull the full bug memory from Mentu, including body, meta, and any attachments.
Investigate — Read relevant source files, trace the bug to its root cause, and identify the files that need to change.
Estimate Complexity — Classify the fix into a tier:
- T1 — Simple: CSS, text, config, or single-line logic changes.
- T2 — Moderate: multi-line logic, component refactors, or state changes.
- T3 — Complex: multi-file changes, schema migrations, or cross-cutting concerns.
Branch — Create a git branch named fix/<mem_id_short>-<slug> from the current HEAD.
HANDOFF — Write a HANDOFF.md document in the branch describing the diagnosis, fix plan, and affected files. This serves as a breadcrumb if the agent is interrupted.
Commit + Claim — Create a Mentu commitment (commit op) and immediately claim it (claim op). This signals to other agents that the ticket is being worked on.
Fix — Apply the code changes. Record progress evidence on each significant step.
Verify — Run the build command and capture the result. If the build fails, attempt one retry. A passing build is required to proceed.
Push + PR — Push the branch and create a pull request via gh pr create. Capture the PR URL as evidence.
Submit — Close the commitment with pass status and the full evidence array.

Example

/fix mem_a8f3c21d

Fetching memory mem_a8f3c21d...
  Title: Login form rejects valid email
  Severity: high

Investigating...
  Root cause: email regex in src/lib/validators.ts rejects "+" in local part
  Files: src/lib/validators.ts, src/components/LoginForm.tsx
  Tier: T1 (single regex fix)

Creating branch fix/a8f3c21d-login-email-regex...
Writing HANDOFF.md...
Committing to Mentu (mem → commitment)...
Claiming commitment...

Fixing...
  ✓ Updated EMAIL_REGEX in src/lib/validators.ts
  ✓ Added test case for "user+tag@example.com"

Verifying...
  ✓ Build passed (3.2s)

Pushing & creating PR...
  ✓ PR #142: "fix: accept + in email local part"
    https://github.com/org/repo/pull/142

Submitting to Mentu...
  ✓ Commitment closed (pass)
  Evidence: 4 items (commit, build, PR, validation)

Done. Fix shipped in 47s.

/run

Full autopilot mode. Runs waves of triage + fix cycles. Each wave triages the backlog, picks the top N tickets, fixes them, pushes all results, and summarizes before continuing to the next wave.

Syntax

/run [flags]

Flags

Flag	Default	Description
`--dry-run`	`false`	Triage and plan only. No branches, no commits, no PRs.
`--max-waves N`	`5`	Maximum number of waves before stopping.
`--batch-size N`	`5`	Number of tickets to fix per wave.

What It Does

Wave Start — Run /triage to get the current backlog.
Batch Fix — Take the top N tickets (by score) and run /fix on each sequentially.
Push All — Ensure all branches are pushed and PRs are open.
Wave Summary — Log a summary of what was fixed, failed, or skipped.
Circuit Breaker Check:
- If 2 consecutive waves produce zero fixes, stop (backlog is empty or all remaining tickets are unfixable).
- If max-waves is reached, stop.
Continue — If no breaker tripped, start the next wave.

Example

/run --max-waves 3 --batch-size 3

Wave 1/3
  Triaged: 12 → 5 actionable
  Fixing: mem_a8f..., mem_b4d..., mem_c1a...
    ✓ mem_a8f — PR #142
    ✓ mem_b4d — PR #143
    ✗ mem_c1a — build failed (submitted as fail)
  Wave 1 complete: 2 passed, 1 failed

Wave 2/3
  Triaged: 9 → 4 actionable
  Fixing: mem_d7e..., mem_e9c..., mem_f0b...
    ✓ mem_d7e — PR #144
    ✓ mem_e9c — PR #145
    ✓ mem_f0b — PR #146
  Wave 2 complete: 3 passed, 0 failed

Wave 3/3
  Triaged: 6 → 2 actionable
  Fixing: mem_g2h..., mem_h3i...
    ✓ mem_g2h — PR #147
    ✓ mem_h3i — PR #148
  Wave 3 complete: 2 passed, 0 failed

Autopilot finished: 3 waves, 7 passed, 1 failed, 8 PRs opened.

Dry Run

/run --dry-run

Wave 1 (dry run)
  Triaged: 12 → 5 actionable
  Would fix: mem_a8f (T1), mem_b4d (T2), mem_c1a (T1), mem_d7e (T2), mem_e9c (T3)
  Estimated effort: 3× T1, 1× T2, 1× T3

No mutations made. Run without --dry-run to execute.

/setup

Interactive onboarding. Walks you through configuring the Mentu connection for the current project.

Syntax

/setup

What It Does

Prompts for the Mentu API URL (provides default: https://mentu-proxy.affihub.workers.dev/ops).
Prompts for your API token.
Prompts for your workspace ID.
Tests the connection by making a lightweight API call.
On success, writes .mentu.json to the project root with the validated config.
Adds .mentu.json to .gitignore if not already listed.
Displays a confirmation with workspace name and memory count.

Example

/setup

Mentu Autopilot Setup
━━━━━━━━━━━━━━━━━━━━━

API URL [https://mentu-proxy.affihub.workers.dev/ops]: ↵
API Token: ••••••••••••••••
Workspace ID: a1b2c3d4-e5f6-7890-abcd-ef1234567890

Testing connection...
  ✓ Connected to workspace "my-project"
  ✓ 14 open memories found

Wrote .mentu.json
Added .mentu.json to .gitignore

Setup complete. Run /triage to see your backlog.

/status

Quick pipeline overview. Shows commitment counts, recent activity, and throughput metrics for the connected workspace.

Syntax

/status

What It Shows

Connection status — whether the plugin can reach the Mentu API.
Commitment counts — how many commitments are open, claimed, and closed.
Recent activity — submissions and failures in the last 24 hours.
Throughput — rolling 7-day average of fixes per day.

Example Output

Mentu Pipeline Status
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Workspace:    my-project (a1b2c3d4-...)
Connection:   ✓ Connected

Commitments:
  Open:       3
  Claimed:    1
  Closed:     27

Recent Activity (last 24h):
  Submitted:  4
  Failed:     0

Throughput:   4.2 fixes/day (7-day avg)

If the connection fails, the output will show the error and suggest running /setup to reconfigure.