Skip to content

Vivary guides

Use these Vivary guides to complete one governed workspace task at a time. Each page is the canonical human-and-agent procedure that the public site renders. The guides use STE100 style. Each procedure is concise, direct, and safe to copy.

The complete guide set is thorough. Each guide keeps one task boundary. The command reference keeps exhaustive flags, schemas, limits, and exit codes.

The guide library describes published 0.4.2. Registry latest installs it. Use the release-status table as the publication authority.

The historical proof records the published 0.3.1 full layout. Do not use that 38-file fixture as the thin-workspace expectation.

Run each guide with the published launchers. Use these commands:

Terminal window
uvx create-vivary
uvx --from vivary-tropo tropo

The guide commands use these launchers. Replace each example workspace path with the intended absolute path.

Install the tools first if you prefer bare create-vivary and tropo commands. Get approval before you install anything into a shared environment.

Install into an environment on Windows:

Terminal window
python -m venv C:/path/to/vivary-env
C:/path/to/vivary-env/Scripts/python.exe -m pip install vivary
C:/path/to/vivary-env/Scripts/Activate.ps1

Install into an environment on macOS or Linux:

Terminal window
python -m venv /path/to/vivary-env
/path/to/vivary-env/bin/python -m pip install vivary
source /path/to/vivary-env/bin/activate

The vivary meta package installs the CLI suite. Use it for every --governed or record command. Pin create-vivary==0.3.1 only to get the historical full layout.

Task Guide Result
Start a new project Create a Vivary workspace Create the five-file seed.
Give an agent context Connect an agent to Vivary Use the standard route or optional MCP.
Retrieve evidence Get bounded context Return task context or a Task Capsule.
Preserve earned context Write one approved record Apply one capsule-bound record.
Add Vivary to a project Adopt an existing project Apply a bounded brownfield plan.
Prove health Verify and recover a workspace Validate health and use explicit recovery.

For a new project:

  1. Create the workspace.
  2. Verify the five-file seed.
  3. Connect the selected agent.
  4. Retrieve bounded context.
  5. Complete real work.
  6. Verify the result.
  7. Write one record only when the work earns it.

For an existing project:

  1. Preview adoption.
  2. Review conflicts and privacy.
  3. Get approval for the exact plan hash.
  4. Apply the approved plan.
  5. Verify the adopted workspace.
  6. Connect the selected agent.

Each guide contains these parts:

  • result
  • agent contract
  • required input
  • exact procedure
  • expected output
  • files or state that can change
  • refusal conditions
  • verification
  • recovery
  • links to the canonical reference

Humans and agents use the same canonical guide. The public site renders these Markdown files. The agent indexes expose the same content.

Do not create a second agent manual. Do not copy guide text into workspace seeds. Do not add guide packs during init, adoption, MCP startup, or retrieval.

Default retrieval is read-only. The --receipt option writes a local privacy-preserving receipt. MCP is read-only and optional. Plain Doctor is read-only. Init creates the approved new-workspace seed. Adoption needs an exact approved plan hash. Record apply needs a capsule and exact approved plan hash. Recovery needs the exact reported transaction hash.

Publishing, external writes, destructive work, credentials, and authority expansion remain human gates.