FAQ and glossary
Keep claims tied to the operations you actually observed.
Am I building a practice team or my actual team?
The default creates the team you intend to use for work. Start with one small, useful non-sensitive task. Continue in the same folder when the output passes your checks and a fresh chat recovers saved state. The fictional demo is optional and separate.
What do I do tomorrow?
Open the same project and send resume in a fresh chat. Review saved status and next action. For new work, define a new task and update the loader to point to it. See Use your team in the menu.
How do I open the right folder?
Let the setup prompt use supported app tools first. If it cannot open the project, it must give an exact path and one verified manual step. Reading an attached file does not change the active workspace.
Do I need to code?
No. The task and instructions are readable Markdown. Node 20 or later is needed only to execute the validator. Manual inspection is available, with a different result label.
How long does setup take?
Record real trial time. No duration is promised until beginner trials establish it.
Is this full YAML?
No. The header starts and ends with an exact --- line. Exactly six key: value lines appear in order: id, revision, status, role, mode, next. No unknown keys, duplicates, nesting, lists, quotes, comments, or blank header lines. Values are trimmed single-line text. Revision is a canonical positive safe integer. The shipped validator defines the restricted grammar.
Can the AI access only this folder?
Instructions do not enforce isolation. Observe the explicitly authorized disposable-marker test and record one of its five outcomes. Policy refusal and unavailable access do not prove a sandbox.
Does deleting chat prove the files caused the restart?
No. Saved memory may remain. Verify isolation in a separate profile/account or checked session. A different profile alone may share account memory. Record the controls and their scope, or record restart passed, memory not excluded. Preserve account boundaries.
Does the reply label prove a team of agents ran?
No. The work-team route uses Role: lead | Mode: single_session_role; the optional demo uses research. Both report a separate Session: fresh or resumed line. It is one role in one session. Separate execution belongs in Guide 2.
Will this make Mem unnecessary?
It removes dependence on Mem for task state and team instructions. Mem may still help capture and retrieve notes. Promote accepted team knowledge deliberately, and retain authoritative external business records in their approved systems. Decide from actual retrieval needs.
Are skills and plugins useful?
A plain-file skill helps repeat a procedure and is already included. Add a plugin or connector only when an authorized task needs its tools. Keep portable instructions on disk and product-specific setup in adapters. Verify actual access after installation.
Can I use OneDrive?
Validated non-sensitive plain-file snapshots can be transported by sync. Use one writer. Keep active locks, queues, databases, verbose runtime logs, and Git metadata in a local nonsynced working location. Sync is not locking or tested restoration.
Glossary
- Adapter
- Product-specific setup, capability observations, and handoff guidance around the portable files.
- Context
- The selected information sent to the tool for the current task. Cloud processing receives that context.
- Handoff
- A section in the task naming artifacts, verified work, blockers, and restart instructions. Current state stays in the header.
- Local inference
- A model running on hardware you control. Local files alone do not establish this.
- Role
- A reusable work contract with purpose, authority, evidence, and quality criteria.
- Skill
- A repeatable procedure with inputs, steps, evidence, and done-when criteria.
- Task record
- Human-readable Markdown with one restricted, validated header and four task sections.
- Plugin
- A product package that may add skills, tools, or app connections. Its runtime is product-specific.
- Proof record
- Actual observed operations, results, failures, elapsed time, configuration, and limitations.