GQ-001 · The Agent Workspace

How should the complete agent workspace be designed across code layout, agent architecture, memory, enforcement, and agent-generated user interfaces?

SISO Open Source Foundation

active

partial

Review due · assessed 2026-08-02 · due 2026-09-01

Run one cold reconstruction task against the typed workspace boundary and compare errors, intervention, and elapsed time with the prior evidence path.

gls:release:0212a6f4-1860-453b-bb27-2e9ed70a09c4 · program-metadata-1.0.0+ea7f0e4

Not released — the selected Release contains question-program metadata, not an answer artifact.

hybrid

How to lay out a new agent workspace — what to enforce mechanically, what to leave to convention, and which patterns to refuse to repeat.

Each principle is graded by the evidence behind it and states whether it is a universal finding or a SISO preference — the answer_shape's own distinction, made checkable per principle rather than claimed for the set. · Rejected patterns are recorded with the observation that rejected them, so a later reader can tell a tried-and-failed pattern from an untried one. · Every architectural claim names a workspace where it was actually run, not a design sketch. · A principle that has never been violated in practice is marked as untested rather than as confirmed.

A principle stated as universal is contradicted by a workspace outside SISO, making it a preference that was mislabelled. · The architecture cannot be described without naming a specific tool, meaning what was learned is a vendor workflow rather than a workspace design. · Following the principles produces a workspace that a new agent cannot orient in — the design optimises for its author rather than for a cold reader. · Enforcement mechanisms are routinely bypassed by the agents they govern, indicating the design describes intended rather than actual behaviour.

Evidence comes almost entirely from a single workspace operated by a single agent, so 'universal' cannot yet be distinguished from 'true here'. The answer_shape demands exactly that distinction. · Memory architecture is named in the question but only lightly exercised — no claim rests on measured memory behaviour. · Agent-generated user interfaces have one instance (the observatory) and no comparison, so its design choices are untested against alternatives.

An enforcement mechanism (gate, hook, contract) is added or disabled, which changes what the workspace actually guarantees. · A new agent joins the workspace and takes measurably longer to orient than the design predicts. · A principle is violated in practice without the violation being caught, exposing an unenforced claim. · Agent-generated UI surfaces begin carrying values that are not derived from a named source.

public agent-workspace repositories · code-layout and architectural enforcement · memory and context systems · agent-facing interface patterns

Evidence-graded principles, reference architectures, rejected patterns, and tests that distinguish universal findings from SISO preferences.

Re-answer when a major agent platform, workspace architecture, or failed forward prediction changes a load-bearing conclusion.

public metadata only

A complete agent workspace can integrate its organs through typed interfaces without collapsing their independent ownership boundaries. Scope: agent-workspace architecture · Falsifier: Representative workspace tasks require one organ to own another organ's durable state, evidence adjudication, and runtime authority to remain coherent. · Depends on: none · Evidence: EC-GQ001-RESEARCH-METHOD · Review: 2026-08-02

Stable identities, typed relationships, and explicit receipts reduce the context a cold agent must reconstruct before useful work. Scope: workspace reconstruction · Falsifier: Cold-task comparisons show no meaningful reduction in reconstruction time, errors, or human intervention after adding the addressable contract. · Depends on: QA-GQ001-BOUNDARIES · Evidence: EC-GQ001-LIBRARY-SEED · Review: 2026-08-02

The question-driven research architecture separates identity, discovery, corpus, evidence transformation, and execution responsibilities. Owner: The Great Library of SISO · Supports: QA-GQ001-BOUNDARIES · Challenges: none · Observed: 2026-08-02

The immutable seed exposes publication-safe identity and scope without presenting the partial answer as a public artifact. Owner: The Great Library of SISO · Supports: QA-GQ001-ADDRESSABILITY · Challenges: none · Observed: 2026-08-02

The current question frame defines learning demand but grants no implementation or budget authority. Owner role: question steward · Status: proposed · Recorded: 2026-08-02T00:34:00+07:00 · Truth: not applicable · Predecessors: none · Assumptions: QA-GQ001-BOUNDARIES, QA-GQ001-ADDRESSABILITY