pdurlej/kan-ductor

Fork 0

gemini(cockpit): actor/source badges inventory across cockpit and activity surfaces #134

New issue

Closed

opened 2026-05-28 01:20:44 +02:00 by codex · 1 comment

codex commented

2026-05-28 01:20:44 +02:00

Collaborator

Parent: #2
Agent lane: Gemini 3.5 Flash
Wave: cockpit / read model
Risk class: low

Goal

Make trust metadata visible consistently: who said/did what, from where, and with what authority.

Context refs

Open issues #83, #80
docs/operational-cockpit-contract.md

Scope

Inspect cockpit, recent activity, audit view, and card context surfaces.
Inventory actor/source/channel badges or missing equivalents.
Recommend small consistent vocabulary.

Acceptance

Inventory covers Piotr/Iskra/Codex/system and Signal/Kan/Forgejo/Obsidian sources.
Operator-approved vs tool-result/proposal is not blurred.
No broad design rewrite.

Suggested checks

Static UI/code search.

Non-goals / fences

Do not deploy, restart production, rotate secrets, or run production migrations.
Do not widen MCP write authority or public exposure.
Keep the change small enough for one focused PR or one scouting report.

Expected output

A short PR or issue comment with findings, touched files, tests run, and remaining risks.

Parent: #2 Agent lane: Gemini 3.5 Flash Wave: cockpit / read model Risk class: low ## Goal Make trust metadata visible consistently: who said/did what, from where, and with what authority. ## Context refs - Open issues #83, #80 - `docs/operational-cockpit-contract.md` ## Scope - Inspect cockpit, recent activity, audit view, and card context surfaces. - Inventory actor/source/channel badges or missing equivalents. - Recommend small consistent vocabulary. ## Acceptance - Inventory covers Piotr/Iskra/Codex/system and Signal/Kan/Forgejo/Obsidian sources. - Operator-approved vs tool-result/proposal is not blurred. - No broad design rewrite. ## Suggested checks - Static UI/code search. ## Non-goals / fences - Do not deploy, restart production, rotate secrets, or run production migrations. - Do not widen MCP write authority or public exposure. - Keep the change small enough for one focused PR or one scouting report. ## Expected output A short PR or issue comment with findings, touched files, tests run, and remaining risks.

codex added the

labels

2026-05-28 01:20:44 +02:00

codex commented

2026-06-03 09:15:34 +02:00

Author

Collaborator

Codex scouting pass after #140 / before badge UI work.

Inventory

Card activity surface

File: apps/web/src/views/card/components/ActivityList.tsx

Already good:

Activity uses agentAttribution.displayActorName before falling back to the Kan user.
Agent actions can show a small via agent / via automation badge.
API schema exposes agentAttribution.auditPublicId, actor, displayActorName, machineIdentityPublicId, machineIdentitySlug, machineIdentityDisplayName, machineIdentityKind in packages/api/src/schemas/card.ts.

Gap:

Badge vocabulary is generic (via agent) and does not show source/channel (Signal, MCP, n8n, etc.).
It does not expose audit id or confidence inline, which is probably fine for card view, but useful as tooltip/follow-up.

Agent audit surface

File: apps/web/src/views/settings/AgentAuditSettings.tsx

Already good:

Actor display prefers machineIdentityDisplayName ?? machineIdentitySlug ?? actor.
Shows mode, action, date, source badge, raw/redacted payload badge, audit id, optional idempotency key, optional source ref, identity slug.
Source filters include signal, fastmail, meeting, forgejo, manual, obsidian, iskra, n8n.

Gap:

Source/actor badges are local to audit view; no shared vocabulary/component yet.
machineIdentityKind is rendered as plain text, not a consistent badge.

Agent proposals surface

File: apps/web/src/views/settings/AgentProposalsSettings.tsx

Already good:

Shows proposal status.
Shows Prepared by <createdBy>.
Extracts sources from create_card action metadata and renders them as a compact source chip.

Gap:

Source chip only looks at create_card metadata today.
It does not distinguish operator-approved, tool-result, proposal, applied, or rejected as trust states beyond status.

Operational cockpit surface

File: apps/web/src/views/settings/OperationalCockpitSettings.tsx

Already good:

Shows sourceSystem, domain/status/risk/needsPiotr/stale, evidence kind/ref.
It is suitable for producer-side trust metadata via snapshot items.

Gap:

No explicit actor badge because snapshot items are evidence/read-model entries, not necessarily actions.
Should avoid implying “who did it” unless producer supplies a safe actor/source field later.
#133/#141 addresses risky evidence refs in the default UI.

Recent activity / MCP read path

Docs: docs/agent-mcp-contract.md

Already good:

Contract names actorKind, displayActorName, viaAgentTokenName, agentAuditPublicId, agentAttributionConfidence, list public IDs/names.
Contract explicitly says Iskra actions should display as OpenClaw Iskra MCP, not Piotr.
agentAudit[] includes machine identity fields.

Gap:

This is strong in API/MCP contract, but not consistently surfaced as one shared UI badge vocabulary.

Recommended small next PR

Create a shared UI helper/component for trust chips, without redesigning pages:

ActorChip: human / agent / automation / system, display name, optional machine identity kind.
SourceChip: signal / fastmail / meeting / forgejo / manual / obsidian / iskra / n8n / kan / mcp.
TrustStateChip: proposal / approved / applied / rejected / audit-exact / audit-low-confidence / redacted.

Use it first in AgentAuditSettings and maybe ActivityList; do not touch every surface in one PR.

Vocabulary proposal

Actors:

Piotr / human
OpenClaw Iskra MCP / agent
codex-<thread-id> / agent
n8n-<workflow> / automation
system / system

Sources:

Signal, Kan, Forgejo, Obsidian, Fastmail, Meeting, Iskra, n8n, MCP

Trust states:

manual
via agent
proposal
approved
applied
audit exact
audit low confidence
redacted

Non-goals confirmed

No production change, no token/identity change, no MCP authority change. This was static code/docs scouting only.

Codex scouting pass after #140 / before badge UI work. ## Inventory ### Card activity surface File: `apps/web/src/views/card/components/ActivityList.tsx` Already good: - Activity uses `agentAttribution.displayActorName` before falling back to the Kan user. - Agent actions can show a small `via agent` / `via automation` badge. - API schema exposes `agentAttribution.auditPublicId`, `actor`, `displayActorName`, `machineIdentityPublicId`, `machineIdentitySlug`, `machineIdentityDisplayName`, `machineIdentityKind` in `packages/api/src/schemas/card.ts`. Gap: - Badge vocabulary is generic (`via agent`) and does not show source/channel (`Signal`, `MCP`, `n8n`, etc.). - It does not expose audit id or confidence inline, which is probably fine for card view, but useful as tooltip/follow-up. ### Agent audit surface File: `apps/web/src/views/settings/AgentAuditSettings.tsx` Already good: - Actor display prefers `machineIdentityDisplayName ?? machineIdentitySlug ?? actor`. - Shows `mode`, action, date, source badge, raw/redacted payload badge, audit id, optional idempotency key, optional source ref, identity slug. - Source filters include `signal`, `fastmail`, `meeting`, `forgejo`, `manual`, `obsidian`, `iskra`, `n8n`. Gap: - Source/actor badges are local to audit view; no shared vocabulary/component yet. - `machineIdentityKind` is rendered as plain text, not a consistent badge. ### Agent proposals surface File: `apps/web/src/views/settings/AgentProposalsSettings.tsx` Already good: - Shows proposal status. - Shows `Prepared by <createdBy>`. - Extracts sources from `create_card` action metadata and renders them as a compact source chip. Gap: - Source chip only looks at `create_card` metadata today. - It does not distinguish `operator-approved`, `tool-result`, `proposal`, `applied`, or `rejected` as trust states beyond status. ### Operational cockpit surface File: `apps/web/src/views/settings/OperationalCockpitSettings.tsx` Already good: - Shows `sourceSystem`, domain/status/risk/needsPiotr/stale, evidence kind/ref. - It is suitable for producer-side trust metadata via snapshot items. Gap: - No explicit actor badge because snapshot items are evidence/read-model entries, not necessarily actions. - Should avoid implying “who did it” unless producer supplies a safe actor/source field later. - #133/#141 addresses risky evidence refs in the default UI. ### Recent activity / MCP read path Docs: `docs/agent-mcp-contract.md` Already good: - Contract names `actorKind`, `displayActorName`, `viaAgentTokenName`, `agentAuditPublicId`, `agentAttributionConfidence`, list public IDs/names. - Contract explicitly says Iskra actions should display as `OpenClaw Iskra MCP`, not Piotr. - `agentAudit[]` includes machine identity fields. Gap: - This is strong in API/MCP contract, but not consistently surfaced as one shared UI badge vocabulary. ## Recommended small next PR Create a shared UI helper/component for trust chips, without redesigning pages: - `ActorChip`: human / agent / automation / system, display name, optional machine identity kind. - `SourceChip`: signal / fastmail / meeting / forgejo / manual / obsidian / iskra / n8n / kan / mcp. - `TrustStateChip`: proposal / approved / applied / rejected / audit-exact / audit-low-confidence / redacted. Use it first in `AgentAuditSettings` and maybe `ActivityList`; do not touch every surface in one PR. ## Vocabulary proposal Actors: - `Piotr` / human - `OpenClaw Iskra MCP` / agent - `codex-<thread-id>` / agent - `n8n-<workflow>` / automation - `system` / system Sources: - `Signal`, `Kan`, `Forgejo`, `Obsidian`, `Fastmail`, `Meeting`, `Iskra`, `n8n`, `MCP` Trust states: - `manual` - `via agent` - `proposal` - `approved` - `applied` - `audit exact` - `audit low confidence` - `redacted` ## Non-goals confirmed No production change, no token/identity change, no MCP authority change. This was static code/docs scouting only.

codex closed this issue

2026-06-03 09:15:48 +02:00