Try Intent-System→
Proof

Start with a direction. Follow the evidence.

One short direction-setting excerpt from a longer measured report, one traceable implementation chain, and one honest v0.22.0 authoring-only minimum start. The excerpt is not the whole originating prompt; the workflow that follows is structured and reviewable, not automatic or trivial.

Follow the full pipeline ↓See the authoring-only startCurrent v0.22.0 model
Intent-System · Step 2 of 9↑ Overview
  1. 01Current Flow
  2. 02Evidence Showcase
  3. 034-Thread Orchestration
  4. 04Timer-Loop Alternative
  5. 05Intent Tree
  6. 06Unified Intents
  7. 07Human + AI Assignment
  8. 08Review & Closeout Loop
  9. 09Roadmap
Case 01 · full pipeline

A measured report became a traceable implementation slice

This chain starts with the detailed measured report in public issue #1489, follows the G689 contract in public issue #1490, and ends at PR #1492 and its merge. Public issue #1493 (G690) was created after that merge, so it appears only as a later follow-on/corroborating record, never as the origin. Every surrounding explanation below is labeled as a paraphrase unless it is marked as an observed fact or a verbatim excerpt.

  1. 01
    Verbatim excerptOrigin excerpt

    A direction-setting excerpt from the full report

    “And the design thread should be able to act when it notices one, rather than watching a seat sit blocked.”Excerpt from public issue #1489 · G690/#1493 later corroboration

    This sentence is reproduced exactly from the longer measured bug report in public issue #1489. It is a short direction-setting excerpt, not the whole originating prompt. G690/#1493 was created after the #1492 merge and is later corroboration only, never the origin. The Japanese text on this page explains the evidence; it is not presented as an original quotation.

  2. 02
    ParaphraseDesign

    Public issue #1490 names the constraint

    Paraphrase: Public issue #1490 (G689) turns the observed shell-approval gap into a two-layer design. A prompt class recognizes and extracts the dialog; scoped policies decide what may be answered. Compound commands are checked segment by segment and unknowns fail closed.

    Sourcespublic issue #1490 (G689)PR #1492
  3. 03
    ParaphrasePacket

    The implementation packet makes the boundary checkable

    Paraphrase: Public issue #1490 (G689) fixes the three named scopes — `project-test`, `owned-scratch-delete`, and `exact-command-once` — plus prompt-class inspection and bounded per-dialog answers. Answer authority stays orchestration-only in this slice.

    Sourcespublic issue #1490 (G689)PR #1492
  4. 04
    ParaphrasePublic issue

    The packet becomes issue #1490

    Paraphrase: Public issue #1490 carries the standalone G689 contract, its acceptance criteria, verification, and `main` base-branch policy.

    Sourcesissue #1490
  5. 05
    ParaphrasePull request

    PR #1492 carries the implementation

    Paraphrase: PR #1492 records the shell-command prompt class, conservative shell-AST extraction, the three scoped policies, audit fields, and prompt-class list/describe inspection. It targets `main`.

    SourcesPR #1492
  6. 06
    ParaphraseMerge

    The observed endpoint is a merge on main

    Observed fact: PR #1492 merged to `main` at `b80d358913be6375741fe95ef93113159b2e0087`. The evidence ends at this verified merge; no automatic downstream work is inferred from it.

    Sourcesb80d3589 merge commitPR #1492
What this proves

The evidence supports a narrow claim: a recorded request can become a durable design, packet, issue, PR, and merge chain with named boundaries and exact identifiers. It does not support a claim that the workflow is automatic, effortless, or free of human approval.

Case 02 · authoring-only minimum

When the goal is to author and publish an issue only

With `team_mode=authoring-only`, you can shape intent and publish an issue without writing code. It is a preview-through-1.x authoring capability. The mode is orthogonal to the session-layer transport: delivery remains the default, herdr-only remains preferred for collocated teams, and agmsg + herdr remains supported for distributed teams or existing agmsg investments.

Source-verified commands

Record the smaller team shape

intent-cli team-mode set --domain <domain> --team <team> --mode authoring-only --write --format json
intent-cli team-mode show --domain <domain> --team <team> --format json
intent-cli team-mode validate --domain <domain> --team <team> --format json

These are the released v0.22.0 guide commands. The mode record is separate from transport selection.

Sourcesv0.22.0 guide
Authoring front door

The honest minimum path

  1. Shape or interview intent at the front door.
  2. Author a standalone packet.
  3. Pass the existing readiness, claim, repository, content, duplicate, and branch-lane gates.
  4. Publish the issue and record the external handoff.

G691/G692 keep the publish boundary explicit: authoring-only does not silently become delivery.

Sourcespublic issue #1495 (G691)public issue #1497 (G692)PR #1496PR #1498d305987b merge05b0aa57 merge
Ends at issue publication / handoff

Authoring-only records the design actor, explicit operator acceptance, destination ownership, and a `published-external-handoff` record. That makes the handoff observable while preserving the publish gates; it does not authorize a worker or bypass them.

  • No delivery topology or four delivery seats are required.
  • Implementation, review, CI, delegation, supervision, and merge are not part of this minimum-start path.
  • intent-cli emits guidance and verifies state; it does not launch providers, operate herdr, or register schedulers.
Sourcesv0.22.0 guide (EN)v0.22.0 guide (JP)v0.22.0 release notes (EN)v0.22.0 release notes (JP)
Evidence ledger

The source boundary stays visible

The two cases have different origins and different endpoints. The full-pipeline story is a selected measured-to-merged chain. The authoring-only story is a released capability description that stops before delivery work.

Try Intent-System→