Skip to content

Slash commands reference

Every Dezycro command your coding agent runs is a skill — a markdown playbook the agent reads and follows interactively. This page lists each one: what it's for, what it needs, and what it does.

How the commands are named

Only one command is built into the Dezycro CLI: /dezycro:init, which binds a repository to a workspace.

Everything else is a platform skill — Dezycro ships it into every workspace, and it syncs to your machine like any other workspace skill, namespaced by your workspace's prefix and named dz-*:

Workspace prefix Claude Code Cursor
acme /acme:dz-work /acme-dz-work

Two things follow from this:

  • It's always clear which workspace you're acting on. A skill synced from the acme workspace only ever talks to acme; if your repo is bound to a different workspace, the skill stops and tells you.
  • Your team can see, fork, or opt out of any of them on the Skills page — see Platform skills.

The examples below use acme as the prefix — the short slug your workspace admin sets under Workspace settings → Skills Namespace. Substitute your own (shown on the Skills page, or dezycro skills list). Skills sync automatically at session start and every 5 minutes while the gateway runs; to pull them right now, run dezycro skills sync.

You don't have to remember the names

Each skill carries a description of when it applies, so you can just say what you want — "start working on the checkout feature", "run the verifier", "remember that we never use gRPC for new APIs" — and the agent picks the right one.

/dezycro:init

Bind a repository to a Dezycro workspace and project. Run once per repo.

Checks the MCP connection, asks which workspace and project this repo belongs to, detects your dev / staging / prod environments, and writes .dezycro/config.json. Every other command reads that file first.

/acme:dz-work [feature]

Start or resume a coding session on a feature. Run this first in every session.

Accepts a feature name, a substring, or a UUID; with no argument it shows an interactive picker.

  1. Resolves the feature and checks it's workable: the flow graph must be approved (a Draft feature is refused with instructions to approve it in the web app — there's no workbook to work against until then) and both its PRD and TRD must be marked COMPLETE.
  2. Loads the workbook (journeys, tasks, decisions, issues) and the PRD/TRD.
  3. Binds the session to the feature, so every later command and every Companion check takes it as context.
  4. Prints a summary and the operating rules: the workbook is the source of truth, tasks don't go DONE without verifier evidence.
  5. Links every task, decision, and assumption it records to the PRD (S<n> for a section, J<id> for a journey, INFRA for enabling work), lists the feature's open PRD alignment findings, and reconciles any finding sent to it during the session.

/acme:dz-author [feature] [--doc prd | --doc trd] [--revise-existing]

Author or revise a PRD or TRD through a guided flow.

Discovery → clarifying questions → draft → quality gate → finalize. The flow is a state machine that survives context compaction, so you can step away and come back.

  • The TRD quality gate is deterministic and LLM-reviewed: a TRD can't be approved with missing required sections, placeholder tokens (TBD / TODO / FIXME), or intents that aren't grounded in the PRD.
  • Pass --revise-existing to load and edit the current document instead of drafting fresh.
  • Drafting consults your Knowledge Base documents and Agent Playbook pages and cites what it used.

/acme:dz-status

A compact snapshot of the active feature's workbook — journeys and their status, open tasks and issues, recent decisions, the latest verifier result, and the coverage report. Requires an active feature (dz-work first).

/acme:dz-sync

Bring the workbook up to date with what you've actually done. Run it periodically in long sessions, and at wrap-up.

Reviews the conversation and recent commits, then creates tasks for unfinished work, records decisions (with the why), links commits and PRs as artifacts, flags issues you discussed, and notes assumptions. It's also the explicit trigger for Companion's end-of-turn workbook sweep — say "save progress" or run this, and the sweep runs on the next turn end.

/acme:dz-publish-spec

Detect and upload the active feature's OpenAPI spec, which kicks off the build of that feature's verifier binary. Requires an active feature.

  1. Finds the spec — a checked-in file, a build artifact, or a running service's /v3/api-docs.
  2. Validates it and optionally narrows it to the feature's endpoints.
  3. Uploads it. The build runs in the background; dz-verify waits for it.

Run it whenever you add or change endpoints. Companion also nudges you when it sees controller edits since the last publish. What the verifier can check from the published spec — and why a feature can end up with 0 verifiable plans — is in Making your API verifiable.

/acme:dz-verify [--env <name>]

Run the feature's verifier against your local (or a remote) environment. Requires an active feature.

  1. If Companion has flagged unpublished controller changes, runs dz-publish-spec first so you never verify against a stale spec.
  2. Pulls the latest verifier binary (cached by content hash) and waits for a fresh build if one is in progress.
  3. Loads test-persona credentials from .dezycro/auth.local.json (see Personas & Auth).
  4. Runs the binary against the chosen environment's API URL. The binary reports its own results to the workbook — there's no separate upload step, and journey statuses and lifecycle transitions (IN_PROGRESS → VERIFIED) update from that report.
  5. On failures, enters a fix loop: each failing test is triaged first — a real intent violation (the code doesn't do what the PRD says) versus an infra problem, a stale test, or something unclear — then real ones are fixed, and the run repeats — up to five iterations, or until the failing set stops changing. It never claims success it didn't observe.
  6. Shows results and the top coverage gaps, and records the run for Companion's task gating.

If the run reports no verifiable plans, every step returns 401, or steps are skipped, start with the quick diagnosis table.

/acme:dz-import-features

Reverse-engineer an existing codebase into Dezycro features. The fastest way to onboard a legacy repo.

Scans the code for user-facing flows, proposes a feature tree (folders mirror your modules), de-duplicates against features that already exist, then writes a behaviour-focused PRD and a matching TRD for each new feature — for you to review and approve in the web app.

/acme:dz-kb-learn [what was decided]

Record a durable team rule in the Agent Playbook.

The agent invokes it on its own when you state a lasting rule ("from now on…", "we never…", "don't do X again"), when review feedback generalizes beyond the current task, or when a gotcha turns out to be worth keeping. You can also call it directly: /acme:dz-kb-learn use REST, not gRPC, for new service APIs.

It distills a timeless rule (with the why), de-duplicates against existing entries, refuses secrets and task-scoped context, and never blocks your work if the playbook is unavailable.

/acme:dz-kb-recall [topic]

Consult the Agent Playbook before substantive work — a new endpoint, a refactor, a design decision — or on demand: /acme:dz-kb-recall error response format.

Runs a couple of targeted searches, reads the few most relevant pages, and applies what it finds. When the task, the playbook, and the code disagree, it surfaces the conflict instead of guessing.

CLI commands

The skills above run inside your coding agent. The dezycro binary itself has a small command set — see Installation for setup and Workspace Skills for syncing.

Command Purpose
dezycro login / dezycro setup Authenticate, then wire the MCP connection and hooks into your installed agents
dezycro run <agent> Launch your agent through the local gateway — team rules, Companion hooks, token-saving output compaction. --status, --restart, --stop
dezycro skills sync / list / sources Sync workspace and platform skills to this machine
dezycro rules [--mode …] Show or set how team rules reach your agent — see Team rules delivery
dezycro workbook (alias wb) Browse a feature's workbook live in the terminal
dezycro feature [--clear] Show (or clear) the feature this repo is bound to
dezycro await-event Block until a platform event lands — e.g. a verifier build finishing. See Waiting on events
dezycro metrics Tokens used and saved, skills synced, models used
dezycro profiles … Manage connection profiles for multiple Dezycro accounts or environments
dezycro update Update the CLI to the latest release

Run dezycro with no arguments on a terminal for an interactive menu.