feat: add durable agent activity preambles by ragesaq · Pull Request #27 · openclaw/clickclack

ragesaq · 2026-06-15T21:35:22Z

Summary

Persist bot-authored commentary and tool activity as durable channel/DM messages, grouped by bot and turn into one collapsible preamble above the final answer.
Keep live agent-responding state separate from history; stale live state expires, while durable preambles survive reloads.
Add independent commentary/tool visibility preferences and configurable self-message alignment.
Document the API-first integration model and regenerate the OpenAPI TypeScript surface.

This keeps the strongest bounded part of the contributor proposal. The unrelated model/reasoning runtime-control plane was removed: ClickClack transports and renders agent activity, while LLM runtime selection remains owned by the external agent runtime.

Security and invariants

New agent_activity:write bot-token scope; deliberately excluded from default bot scope bundles.
Human sessions and bot tokens without the explicit scope cannot create activity rows.
Activity is supported for channels and DMs, not thread replies.
Activity rows do not create unread counts, search results, or browser/server notifications.
Nonce replay identity includes message kind and turn ID in SQLite and Postgres.
Coalescing keys include conversation, author, and turn so separate bots cannot merge activity.

Verification

pnpm check
go test ./apps/api/internal/store/sqlite ./apps/api/internal/store/postgres ./apps/api/internal/httpapi
CI=1 pnpm exec playwright test --reporter=line --workers=1 — 41/41 passed
Production build rendered in the existing Chrome profile; collapsed/expanded preamble and tool details verified
Autoreview clean with no accepted/actionable findings

Thanks @ragesaq for the original agent-activity design and implementation.

Add a message `kind` discriminator ('message', 'agent_commentary', 'agent_tool') and a nullable `turn_id` correlation column so a bot bridge can persist agent commentary/tool output in the normal message stream instead of ephemeral frames. Activity rows ride the ordinary message.created fan-out and inherit channel_seq for scrollback, but are excluded from full-text search (sqlite FTS triggers gated on kind='message'; postgres partial GIN index + search query filter) and from unread/notification accounting (unread subqueries and push notify path skip activity kinds). Creating a message with kind != 'message' requires a bot token carrying the new agent_activity:write scope. The scope is registered in both bot stores' allowed-scope validation but deliberately EXCLUDED from the bot:* bundles, so existing deployments gain no new capability. Human sessions get 403, bots without the scope get 403, unknown kinds get 400. Targets remain single-channel/single-DM via the existing create endpoints. migrations: sqlite 0017, postgres 0011. Tests: authz matrix (403 human, 403 missing scope, 400 unknown kind, success for scoped bot, bundle-exclusion, membership privacy) and store round-trip + unread exemption + search exclusion.

Render durable agent activity messages (kind agent_commentary / agent_tool) inline in the normal message flow with a theme-tied accent rail and a small commentary/tool badge so operators can tell intermediate output from the final answer. The accent comes from a new --activity-accent token (blue on light theme, amber on dark) driven by the existing prefers-color-scheme override, not hardcoded hex. Add a client-only "Show agent activity" preference (default on) in account settings that filters activity kinds from the rendered stream, persisted in localStorage like other client prefs. The client unread bump path also skips activity kinds to match server accounting.

Replace the flat one-row-per-activity rendering (a badged row per agent_commentary/agent_tool message) with a single coalesced preamble block per agent turn, matching the clickglass guidestone: - Consecutive same-turn activity rows (grouped by turn_id) collapse into one block: incrementing commentary prose plus a collapsed-by-default, expandable list of tool calls. Done entirely at render time; no schema or bridge change (rows already carry turn_id). - The block opens while the turn is live (pulsing LIVE pill) and collapses to a single expandable line once the turn ends (DONE). - Tool sub-items render as tool name + derived action + truncated arg detail (one row per call), not the full coalesced step chain. - Replace the single show-agent-activity toggle with two independent operator flags: hide-commentary and hide-tool-calls (localStorage, with migration from the legacy key). Both set omits the block. New: lib/chat/agent-activity.ts (coalescer), lib/chat/tool-detail.ts (glyph/action resolver), components/messages/PreambleBlock.svelte. Also embed dist with all: so underscore-prefixed SvelteKit chunks are not silently dropped from the binary.

Rework the coalesced agent preamble block to the clickglass display contract, per operator review: - Commentary prose and tool calls now interleave in arrival order (commentary, tool, commentary, tool) instead of prose-then-tool-list. - Each tool call is its own row, collapsed by default to one line (glyph + action + tool name + truncated arg detail) on a light blue wash; clicking expands it to the full stored body (step-chain head rendered as plain monospace, output as markdown). - Grouping is by turn_id across the whole list, not adjacency: the bridge's debounced trailing commentary can land after the final answer, which previously split the block and visually swallowed the final message. The final answer now always renders as a clearly separate ordinary message below the block. - The block collapses to a single line when the final answer lands (same-author ordinary message after the turn opened), when anything follows the turn's last activity row, or when the turn goes stale (3 min without frames) so orphaned turns don't pulse LIVE forever. - Preamble wash lightened; tool rows use new --tool-call-bg tokens (light blue on both themes). Both operator flags re-verified across all four states; hide-both omits blocks entirely. Embedded web dist rebuilt.

…collapsed) The pill previously read "done" for any finished turn regardless of whether the block was open. Now it mirrors what the toggle will do: "collapsed" when closed, "expanded" when open, "live" while streaming. Includes embedded dist rebuild.

Within an agent message group the synthetic preamble row is immediately followed by the same author final answer. Tag the paired rows (before-final-message / after-preamble) and draw them as a single bordered card with a flat internal seam, mirroring the ClawCanvas inline model so the activity log and the answer read as one unit. Scoped to the paired rows only; a standalone or still-live preamble keeps its full-bleed wash and gutter accent.

ragesaq · 2026-06-15T21:35:38Z

@clawsweeper this is the durable follow-up to #24 (now merged). It turns the ephemeral activity strip into a persistent, collapsible preamble block: coalesced commentary + per-tool rows (detail collapsed by default), visibility-state pill, and the preamble boxed together with the final answer into one cohesive card, plus a profile toggle.

Green gate on post-merge main: web typecheck 0, build:web 0, embed 0, go build ./... 0, go test -run AgentActivity|AgentProgress PASS. No e2e removal, no lockfile churn, dist regenerates reproducibly.

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds support for durable “agent activity” messages (commentary + tool calls) that are persisted server-side, authorized via a dedicated bot scope, and rendered client-side as a coalesced “preamble block” UI.

Changes:

Introduces kind + turn_id message fields, DB migrations (SQLite + Postgres), and scope-gated HTTP create paths for agent activity rows.
Adds a client-side coalescing pipeline to collapse per-turn agent activity rows into a single preamble block with per-tool expand/collapse UI.
Updates styling/theme variables and fixes Go embed behavior to include underscore/dot-prefixed asset files.

Reviewed changes

Copilot reviewed 31 out of 48 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
apps/web/src/styles/messages.css	Adds styling for the new preamble block and tool-call subrows.
apps/web/src/styles/base.css	Introduces theme variables for activity accents and tool-call row backgrounds.
apps/web/src/lib/types.ts	Extends message/event types to include `kind`, `turn_id`, and client-only `preamble_block`.
apps/web/src/lib/chat/tool-detail.ts	Adds tool name → glyph/action/detail mapping for preamble tool summaries.
apps/web/src/lib/chat/agent-activity.ts	Implements coalescing of agent activity rows into a single synthetic preamble message per turn.
apps/web/src/components/profile/ProfileSettingsModal.svelte	Adds user toggles to hide commentary and/or tool calls.
apps/web/src/components/messages/PreambleBlock.svelte	New component rendering the collapsible preamble + per-tool expansion.
apps/web/src/components/messages/MessageRow.svelte	Renders `PreambleBlock` when `preamble_block` is present and adds pairing classes.
apps/web/src/components/messages/MessageGroup.svelte	Passes adjacent messages into `MessageRow` for preamble/answer pairing logic.
apps/web/src/ChatApp.svelte	Applies coalescing to the message stream and persists hide flags in localStorage.
apps/api/internal/webassets/webassets.go	Switches to `//go:embed all:dist` to include underscore/dot-prefixed asset files.
apps/api/internal/webassets/dist/index.html	Updates built web asset references for the new build output.
apps/api/internal/webassets/dist/200.html	Updates built web asset references for the new build output.
apps/api/internal/webassets/dist/_app/immutable/nodes/5.kueAycLe.js	Build artifact updated for new chunk references.
apps/api/internal/webassets/dist/_app/immutable/nodes/4.Z19KiFas.js	Build artifact updated for new chunk references.
apps/api/internal/webassets/dist/_app/immutable/nodes/3.C3tghYkn.js	Removes old hashed build artifact.
apps/api/internal/webassets/dist/_app/immutable/nodes/3.B-d9fl0n.js	Adds new hashed build artifact.
apps/api/internal/webassets/dist/_app/immutable/nodes/2.BQWrx22g.js	Build artifact updated for new chunk references.
apps/api/internal/webassets/dist/_app/immutable/entry/app.Xpli9ePf.js	Removes old hashed build artifact (entry).
apps/api/internal/webassets/dist/_app/immutable/assets/0.DeZxIIxo.css	Removes old hashed build artifact (CSS).
apps/api/internal/store/types.go	Adds message kind constants, normalization/validation, and new `agent_activity:write` scope.
apps/api/internal/store/sqlite/sqlite.go	Stores normalized kind + turn_id and includes them in realtime event payloads.
apps/api/internal/store/sqlite/sqlc/schema.sql	Adds `kind` and `turn_id` columns to SQLite messages schema.
apps/api/internal/store/sqlite/sqlc/queries.sql	Excludes activity kinds from unread counts and inserts kind/turn_id on message writes.
apps/api/internal/store/sqlite/migrations/0017_agent_activity.sql	SQLite migration adding columns, index, and FTS trigger filtering for kind=message.
apps/api/internal/store/sqlite/helpers.go	Includes `kind` and `turn_id` in message selects/scans.
apps/api/internal/store/sqlite/dms.go	Stores normalized kind + turn_id for direct messages and includes them in events.
apps/api/internal/store/sqlite/bots.go	Makes new scope grantable but not inherited by bot bundles.
apps/api/internal/store/sqlite/agent_activity_test.go	Adds SQLite tests for kind/turn_id round-trip, unread exemption, and search exclusion.
apps/api/internal/store/postgres/sqlc/schema.sql	Adds `kind`/`turn_id` columns and makes search index partial on kind=message.
apps/api/internal/store/postgres/sqlc/queries.sql	Excludes activity kinds from unread counts and inserts kind/turn_id on message writes.
apps/api/internal/store/postgres/search.go	Filters search results to kind=message.
apps/api/internal/store/postgres/postgres.go	Stores normalized kind + turn_id and includes them in realtime event payloads.
apps/api/internal/store/postgres/migrations/0011_agent_activity.sql	Postgres migration adding columns/index and rebuilding partial search index.
apps/api/internal/store/postgres/helpers.go	Includes `kind` and `turn_id` in message selects/scans.
apps/api/internal/store/postgres/dms.go	Stores normalized kind + turn_id for direct messages and includes them in events.
apps/api/internal/store/postgres/bots.go	Makes new scope grantable but not inherited by bot bundles.
apps/api/internal/httpapi/server.go	Adds kind/turn_id to create-message requests and enforces activity scope for activity kinds.
apps/api/internal/httpapi/features.go	Adds kind/turn_id to create-DM requests and enforces activity scope for activity kinds.
apps/api/internal/httpapi/agent_activity_test.go	Adds API-level tests for authz, privacy, and non-inheritance of the new scope.

Files not reviewed (4)

apps/api/internal/store/postgres/storedb/models.go: Generated file
apps/api/internal/store/postgres/storedb/queries.sql.go: Generated file
apps/api/internal/store/sqlite/storedb/models.go: Generated file
apps/api/internal/store/sqlite/storedb/queries.sql.go: Generated file

Comments suppressed due to low confidence (4)

apps/web/src/components/messages/PreambleBlock.svelte:1

class:open is missing its boolean expression, which is inconsistent with the rest of the file and will fail to compile in Svelte versions that require class:name={expr}. Bind it to the open boolean (e.g., class:open={open}) so the chevron rotates only when expanded.
apps/web/src/components/messages/MessageRow.svelte:1
The is-preamble-collapsed class name implies “currently collapsed”, but it is being driven by block.final. Final blocks can be manually expanded in PreambleBlock, so this CSS class will be applied even when the preamble is open, causing “collapsed” spacing/styling to be incorrectly applied. Consider moving the collapsed/expanded styling decisions into PreambleBlock.svelte (where preambleOpen is known) by toggling a class there, or by exposing/binding the open state to the parent so MessageRow can reflect the actual UI state.
apps/web/src/lib/chat/agent-activity.ts:1
Finality detection can become O(T * N) due to scanning from firstIndex+1 to the end for every turn. On long timelines with many turns, this can noticeably impact scroll/render performance. A more efficient approach is a single reverse pass to precompute “is there a later same-author ordinary message after index i” (e.g., track seenOrdinaryByAuthor while iterating from the end), or compute per-index “next ordinary message author” so each turn checks finality in O(1).
apps/web/src/lib/types.ts:1
EventPayload.kind is typed as a plain string, even though Message.kind is already a constrained union ("message" | "agent_commentary" | "agent_tool"). Typing EventPayload.kind to reuse the same union (or Message["kind"]) would prevent typos and keep client logic like handleUnreadBump aligned with server-supported values.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

The preamble now reads clearly yellow while the turn streams (growing bubble), and collapses to a distinct yellow bar capping the blue final answer block. Retunes --activity-accent to yellow on both themes (#d49415 light, #e8b23e dark), strengthens the streaming wash to 17%, and gives the collapsed boxed-pair bar a 22% wash + warm bottom divider so it stands apart from the blue delivery. Rebuilds embedded web dist.

…wcanvas) Per-channel runtime snapshot (model, thinking, context used/limit) served from clickclack's own SQLite/Postgres store, fed by the agent-bridge reading the gateway directly. No /api/live proxy, no ClawCanvas dependency. - migration: channel_runtime table (sqlite 0018 + postgres 0012) - store: GetChannelRuntime / UpsertChannelRuntime / SetChannelRuntimeOverride with disjoint bridge-snapshot vs picker-override columns (neither clobbers the other), raw SQL in both backends - api: GET (session read) / PUT (bot agent_activity:write stamp) / PATCH (session override) /api/channels/{id}/runtime - test: snapshot/override separation + empty-record-on-unknown-channel - docs: RUNTIME_CONTROLS_INDEPENDENT_PLANE.md (data plane + per-field sourcing) Increment 1 of 4. Next: bridge stamp, frontend port (picker+meter), deploy.

…point Increment 2/3 of the composer runtime controls: port the model picker and context meter into the composer utility bar, fed by clickclack's own /api/channels/{id}/runtime endpoint (the increment-1 data plane), with no ClawCanvas dependency. - channel-runtime.ts: runtime resolver + provider/model catalogs (frontend only) - channel-runtime-store.svelte.ts: narrow runtime store; loadChannelRuntime() reads the native endpoint, persistRuntimeOverride() PATCHes picker picks, seedDemoChannelRuntime() keeps the chrome populated until the bridge stamp lands - ComposerModelPicker / ModelPickerMenu / ComposerContextMeter: ported components - ChatComposer: composer-utility row hosts the meter + picker (showModelPicker) - ChatApp: seed on boot, load on channel switch, override handlers persist via PATCH - rebuilt embedded web dist Demoed live on the :8100 stock instance against a real runtime row; meter/picker update end-to-end when the DB row changes. PR intentionally not opened yet.

Bundle the remaining composer/runtime-controls UI work onto the PR: - User (self) messages render in their own accent-tinted bubble that mirrors the agent delivery bubble's weight (7x11 padding, 11px radius, 3px rail) using a single theme-tied token (--user-bubble-soft = color-mix of --accent), so the dark theme override flows automatically. Bubble hugs content (fit-content, capped) so alignment is visible. - Configurable self-message alignment (left default / right) via a new "Your message alignment" select in Account settings. Persisted in localStorage (clickclack:user-align:v1) and applied as a data-user-align root attribute; messages.css mirrors the self group (avatar gutter, header, bubble) without prop drilling. - MessageGroup gains currentUserID -> is-self; bots never match the current user id, so an agent group is never marked self even in a DM. - AgentResponding: compact "agent is responding" status above the composer, driven by any unfinalized tracked turn line. - Context-meter detail popover + model-picker refinements. - Drop the dead "Preamble" title (markup + .preamble-title CSS); the preamble header now reads chevron + state pill + action only.

…ontrols # Conflicts: # apps/api/internal/webassets/dist/200.html # apps/api/internal/webassets/dist/_app/immutable/entry/app.Dphjwod_.js # apps/api/internal/webassets/dist/_app/immutable/entry/app.Xpli9ePf.js # apps/api/internal/webassets/dist/_app/immutable/entry/app.inV1wtz9.js # apps/api/internal/webassets/dist/_app/immutable/nodes/2.BQWrx22g.js # apps/api/internal/webassets/dist/_app/immutable/nodes/2.C6_wREb2.js # apps/api/internal/webassets/dist/_app/immutable/nodes/2.DF-TCSVN.js # apps/api/internal/webassets/dist/_app/immutable/nodes/4.CsfpOieR.js # apps/api/internal/webassets/dist/_app/immutable/nodes/5.BR4pOwJp.js # apps/api/internal/webassets/dist/_app/immutable/nodes/5.ClAa746V.js # apps/api/internal/webassets/dist/_app/immutable/nodes/5.kueAycLe.js # apps/api/internal/webassets/dist/index.html # apps/web/src/ChatApp.svelte # apps/web/src/components/profile/ProfileSettingsModal.svelte

On a ~360px viewport the three full-width runtime widgets (markdown toolbar, context meter, model picker) shared one fixed 1fr/auto/1fr grid and overlapped into an unreadable jumble. Apply the learned mobile treatment below 520px: - composer-runtime-controls drops to two rows: context meter + model picker on top (space-between), markdown glyph strip beneath. - ContextMeter pill collapses to the fill bar + percentage + caret; token counts and cache-hit detail move to the one-tap inspector. - ModelPicker pill collapses to dot + model name (truncated) + caret; provider prefix and thinking/speed labels live in the picker menu. - Right-adjusted self bubbles use the 44px mobile avatar gutter instead of the desktop 56px column for list consistency. Rebuilt embedded web dist.

ragesaq · 2026-06-16T04:43:56Z

Live demo is up in the PR description. Recaptured against the stock validation instance so the chat actually moves: the composer types and sends, the agent works live in-channel (commentary + per-tool rows stream in), the trail folds into the real COLLAPSED gold preamble pill, then expand → collapse, plus the model picker / context meter and a mobile pass. No caption overlays, no frozen frames.

Full-quality MP4 (1600×1000 · 30 fps · 83 s): https://github.com/ragesaq/clickclack/raw/media/composer-demo-video/.proof/video/clickclack-composer-demo.mp4

clawsweeper · 2026-06-16T05:02:39Z

Codex review: needs maintainer review before merge. Reviewed June 23, 2026, 7:53 AM ET / 11:53 UTC.

Summary
This PR adds durable agent-activity preambles, composer runtime controls, self-message bubbles/alignment, channel-runtime storage/API, tests, docs, and rebuilt embedded web assets.

Reproducibility: not applicable. as a bug reproduction: this is a feature PR. The maintainer-review concern is source-verifiable through the runtime PATCH endpoint, generated sqlc merge ref, and accompanying tests.

Review metrics: 3 noteworthy metrics.

Changed files: 83 files affected. The PR spans API, persistent storage, generated sqlc output, web UI, e2e tests, docs, and embedded assets, so maintainers should review it as a broad feature branch.
Runtime routes: 3 routes added. GET, PUT, and PATCH for channel runtime create a new API contract with separate read, bridge-stamp, and member override semantics.
Storage migrations: 4 migrations added. SQLite and Postgres each gain agent-activity and channel-runtime migrations, making upgrade behavior a merge consideration.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🦞 diamond lobster ✨ media proof bonus
Patch quality: 🐚 platinum hermit
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

Get maintainer approval for channel-member model/reasoning overrides.
Optionally rerun final merge-ref sqlc/build/e2e before landing.

Mantis proof suggestion
A real visual pass would materially help confirm the broad chat/composer UI behavior on the merge result. A maintainer can ask Mantis to capture proof by posting this exact PR comment:

@openclaw-mantis visual task: verify durable preamble expand/collapse, self-message alignment, composer model picker, and context meter on the PR merge result.

Risk before merge

[P1] Merging intentionally expands persistent storage/API surface for agent activity and channel runtime state, so maintainers should be comfortable with the upgrade path even though the inspected merge ref preserves current-main sqlc changes.
[P1] The runtime PATCH endpoint intentionally lets a channel member with messages:write set model/reasoning overrides that the bridge later consumes; that policy may be right, but it should be explicitly accepted by maintainers before merge.

Maintainer options:

Approve channel-scoped runtime overrides (recommended)
Maintainers can accept that channel members with messages:write may set next-turn model/reasoning overrides while bridge-stamped effective runtime remains bot-only.
Narrow the override permission
If that policy is too broad, gate PATCH behind a stricter role/scope or default-off setting and keep the bridge PUT contract unchanged.
Require final upgrade proof
Before merge, maintainers can ask for one final current-main merge-ref run of sqlc/build/e2e to confirm the persistent migrations and embedded assets match the landing branch.

Next step before merge

[P2] The remaining action is maintainer policy and ownership review, not an autonomous code repair.

Security
Cleared: No concrete security or supply-chain defect was found; the runtime override surface is tracked as a maintainer auth-provider/product risk instead.

Review details

Best possible solution:

Merge after maintainers explicitly accept the channel-scoped runtime override policy and are satisfied with the inspected merge-ref storage/API result; otherwise narrow PATCH authorization or make the override behavior opt-in before landing.

Do we have a high-confidence way to reproduce the issue?

Not applicable as a bug reproduction: this is a feature PR. The maintainer-review concern is source-verifiable through the runtime PATCH endpoint, generated sqlc merge ref, and accompanying tests.

Is this the best way to solve the issue?

Unclear as a product choice: the implementation is technically coherent and tested, but the channel-writer model/reasoning override policy needs maintainer acceptance before it is clearly the best permanent contract.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 9d29bc6356f2.

Label changes

Label justifications:

P2: This is a normal-priority feature PR with broad but bounded web/API/storage impact.
merge-risk: 🚨 compatibility: The PR changes persistent migrations, generated store code, embedded assets, and upgrade-sensitive message/runtime data shapes.
merge-risk: 🚨 auth-provider: The PR adds model/reasoning override routing and bot-scoped runtime/activity authorization surfaces.
rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🦞 diamond lobster and patch quality is 🐚 platinum hermit.
feature: ✨ showcase: ClawSweeper spotlight: unusually compelling feature idea for maintainer attention. Durable agent preambles plus composer runtime controls are a substantial, cohesive agent-chat workflow improvement worth maintainer attention.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (recording): The PR body and comments contain linked live recording proof plus copied terminal/test output for the changed UI/API behavior; local prepared media extraction failed because ffprobe is unavailable.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body and comments contain linked live recording proof plus copied terminal/test output for the changed UI/API behavior; local prepared media extraction failed because ffprobe is unavailable.
proof: 🎥 video: Contributor real behavior proof includes video or recording evidence. The PR body and comments contain linked live recording proof plus copied terminal/test output for the changed UI/API behavior; local prepared media extraction failed because ffprobe is unavailable.

Evidence reviewed

What I checked:

Repository policy applied: AGENTS.md was read fully; its SQL/sqlc guidance is relevant because this PR changes schema/query files plus generated storedb outputs. (AGENTS.md:1, 9d29bc6356f2)
Clean PR merge ref: GitHub exposes a clean merge result f3e35d8 for head f95564a against the current main ref. (f3e35d806470)
Runtime override policy surface: The merge-ref PATCH endpoint allows an authenticated actor with messages:write and channel access to persist model/thinking overrides after input validation. (apps/api/internal/httpapi/runtime.go:139, f3e35d806470)
Runtime override tests: The PR tests bridge-only PUT, member PATCH, CSRF rejection/acceptance, override validation, and snapshot/override separation. (apps/api/internal/httpapi/runtime_test.go:162, f3e35d806470)
Merge-ref sqlc output: The merge ref includes both current-main member_set_key direct-conversation queries and the PR's kind/turn_id message insert fields in generated SQLite sqlc output. (apps/api/internal/store/sqlite/storedb/queries.sql.go:1258, f3e35d806470)
Current-main runtime hardening retained: The merge ref keeps current-main api() behavior for CSRF headers and empty response bodies, and the PR runtime store routes runtime PATCH through that helper. (apps/web/src/lib/api.ts:10, f3e35d806470)

Likely related people:

Peter Steinberger: Current-main blame shows the base HTTP create-message path, migration runner, and much of ChatApp originated in the initial scaffold commit. (role: introduced API/store and web app baseline; confidence: high; commits: e0de6adf9cf0; files: apps/api/internal/httpapi/server.go, apps/api/internal/store/sqlite/sqlite.go, apps/api/internal/store/postgres/postgres.go)
Vincent Koc: Recent current-main commits changed API/store platform features and hardened client runtime edges that overlap the PR merge surface. (role: recent area contributor; confidence: high; commits: 9d29bc6356f2, 95008dc407a1, 31b23105c3b3; files: apps/web/src/lib/api.ts, apps/api/internal/store/types.go, apps/api/internal/httpapi/server.go)
ragesaq: The merged agent.progress web consumer on current main introduced the AgentProgress surface this PR extends into durable preambles. (role: feature predecessor author; confidence: medium; commits: 1b6e5c71b8d0; files: apps/web/src/ChatApp.svelte, apps/web/src/components/messages/AgentProgress.svelte)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

ClawSweeper keeps one durable marker-backed review comment per issue or PR.
Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
Maintainers can also comment @clawsweeper review to request a fresh review only.
Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

…ime handlers Reject a non-empty turn_id on an ordinary 'message' row (400, ErrTurnIDNotAllowed) at the shared resolveMessageKind choke point used by both the channel and direct-message create paths. turn_id correlates agent activity rows only; an ordinary message carrying one violated the documented 'Empty for ordinary messages' contract (Copilot review, server.go:655). Activity kinds (agent_commentary/agent_tool) may still carry a turn_id. Add runtime_test.go covering the previously 0%-covered composer runtime handlers (get/put/patchChannelRuntime) plus the turn_id contract on both create paths. Lifts go coverage 84.3% -> 86.5%, clearing the 85% gate.

ragesaq · 2026-06-16T05:22:34Z

Pushed 4733232 resolving both open follow-ups on this PR.

1. turn_id contract enforcement (Copilot review, server.go:655)
The shared kind/authorization helper (resolveActivityKind → resolveMessageKind) now also validates turn_id: an ordinary kind="message" carrying a non-empty turn_id is rejected with 400 (store.ErrTurnIDNotAllowed); activity kinds (agent_commentary/agent_tool) may carry one. Enforced at the single choke point both createMessage and createDirectMessage already use, so the documented "Empty for ordinary messages" invariant now holds on both paths. Fail-closed (reject) over silent-clear so a client bug surfaces. Review thread resolved.

2. Go coverage gate (the unstable check)
The red check was the coverage gate (84.3% vs 85.0%), not a test failure. The gap was this PR's own composer-runtime handlers — getChannelRuntime/putChannelRuntime/patchChannelRuntime in runtime.go shipped at 0% coverage. New runtime_test.go covers all three (bridge PUT scope gate, picker PATCH override, composer GET, 404 on unknown channel) plus the turn_id contract on channel + DM paths.

Proof (local, go1.26):

go test ./apps/api/... → all ok
coverage gate (CLICKCLACK_GO_COVERAGE_MIN=85 scripts/check-go-coverage.sh) → 84.3% → 86.5%, exit 0
gofmt -l clean, go vet clean

Diff: +1 sentinel error, helper signature extended (turn_id in/out), both call sites, +1 test file. No behavior change for existing valid callers.

Address Clarity review on PR openclaw#27: - turn_id field doc now states both halves of the contract: must be empty for ordinary messages (kind=message) with a 400 ErrTurnIDNotAllowed, optional for agent activity kinds. - TestMessageTurnIDContract now asserts the 400 response body carries the ErrTurnIDNotAllowed sentinel text (channel + DM paths), proving the failure reason reaches the client rather than an opaque 400.

The composer-runtime-controls work wraps the composer in a .composer-dock that reserves fixed internal chrome above the input card: an always-present agent-responding band plus the runtime-controls row (model picker + context meter). That chrome is intentional dead space, not distance between the newest message and the composer area. expectMessageNearComposer measured the gap to the inner .composer-card, so the new ~28px chrome pushed it past the 24px budget and failed three auto-scroll assertions (unread-divider jump, live-channel pin, escape-to- latest). Re-anchor the measurement to .composer-dock (the composer area boundary) so the assertion stays about "is the newest message pinned to the composer?" rather than tracking composer-internal chrome height, and keep the same 24px slack budget. Falls back to .composer-card when no dock is present. Full Playwright suite: 37 passed.

The history-paging test scrolled to the bottom of a search-jump window and asserted exactly one newer page fetched (toHaveLength(1)). The page count is not an invariant: it depends on how many newer rows fit above the prefetch sentinel before it leaves the trigger margin. A single 50-row page sits right at one viewport, so the agent-activity/composer layout changes on this branch pushed it to the edge where scroll-to-bottom occasionally needs a second sequential page (after_seq=100 then after_seq=150) to fill. That is benign forward fill, confirmed by the sequential cursors, not a runaway to the live tail. This made the test ~37% flaky on this branch (3/8 local repeats) while it is deterministic on main, because the layout changes here moved the fill to the boundary. Assert the real invariant instead: newer-paging engaged (length >= 1) and stayed incremental (history-msg-149 visible, live-tail history-msg-259 still absent). The 259-absence check is the runaway guard. Verified: test 1087 12/12 green after the change (was 3/8); full Playwright suite 37 passed.

Closes the two open HyperReview gates on the runtime cleanup and the stale-carryover bug the new test caught. - web: ComposerContextMeter renders the pill only when the channel reports context-window accounting (context_used/limit). Empty and override-only channels show no meter instead of a meaningless "?/?" placeholder. This also tears the pill down on an in-app channel switch, so the previous channel's meter cannot linger after the picker has already cleared. - api: patchChannelRuntime validates the picker override at the boundary: thinking must be off|minimal|low|medium|high|xhigh|adaptive (empty clears), model is length-bounded (<=256, no control chars). A channel member can no longer persist an arbitrary/oversized string the bridge would apply to a gateway session. Bridge PUT is unchanged (bot + agent_activity:write). - test(e2e): add an in-app sidebar-click channel switch (not page.goto) that asserts the override channel's model chrome and context meter both clear when switching to an empty channel. - test(api): cover cookie+CSRF PATCH (403 without header, 200 with) and the new override validation rejections. - docs: document the meter render-gate and the PATCH validation contract.

ragesaq · 2026-06-16T09:37:45Z

Pushed f95564a resolving the three runtime-control findings from the review, plus the real-behavior proof you asked for.

Demo seed removed (seedDemoChannelRuntime gone) — empty channels render unknown state, not fabricated values.
Stale-clear fixed — store clears before the fetch, guarded by a serial + selected-channel re-check; loadChannelRuntime returns {} for empty channels; transient failure preserves cleared (not stale) chrome.
CSRF write path fixed — persistRuntimeOverride/loadChannelRuntime now route through the api() helper, so a normal cookie session adds X-ClickClack-CSRF: 1 and the override persists.

On top of the review: API-boundary validation on the PATCH (thinking allowlist, model length-bound), and a context-meter render-gate fix for a stale Context ?/? pill that a new in-app (sidebar-click) channel-switch E2E caught.

Proof (network/API behavior):

go test ./apps/api/internal/httpapi/... — PASS: cookie PATCH w/o CSRF → 403, with CSRF → 200 and override_* persisted, bridge/effective fields unchanged, invalid thinking/oversized model rejected.
Full Playwright suite — 38/38 passed, including the in-app channel-switch stale-carryover test.

PR body updated with the full breakdown and the policy note that channel-scoped override (messages:write + membership) is product-approved for this PR.

@clawsweeper re-review

clawsweeper · 2026-06-16T09:37:48Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/27608528771
Updated: 2026-06-16T10:06:25.621Z

steipete · 2026-07-01T07:46:55Z

Maintainer closeout for af57acf4166408ffc19f39bbe8313cd21da8ff10:

Exact-head CI: https://github.com/openclaw/clickclack/actions/runs/28501841719 — Go, TypeScript, Docker Image, and Playwright E2E all passed.
Local gates: pnpm check; focused SQLite/Postgres/HTTP API tests; production E2E 41/41 passed serially.
Rendered proof: fresh production build/server opened in the existing Chrome profile; durable preamble collapse/expand, commentary, tool detail, and final-answer cohesion verified.
Security: activity writes require the dedicated agent_activity:write bot scope; ordinary users and unscoped bots are denied; activity remains outside unread, search, and notification paths.
Product fit: retained durable agent activity and UI preferences; removed the unrelated runtime model/reasoning control plane.
Autoreview: clean, no accepted/actionable findings.

Thanks @ragesaq for the original design and implementation.

ragesaq added 7 commits June 15, 2026 21:25

chore(web): rebuild embedded web dist for consolidated agent activity

c24f2a9

Copilot AI review requested due to automatic review settings June 15, 2026 21:35

Copilot AI reviewed Jun 15, 2026

View reviewed changes

Comment thread apps/api/internal/httpapi/server.go Outdated

ragesaq added 6 commits June 15, 2026 22:13

chore(web): rebuild embedded web dist

7e42cf2

ragesaq changed the title ~~feat(web): durable, cohesive agent activity preamble block~~ feat(web): cohesive agent-activity preamble + composer runtime controls + self-message bubbles Jun 16, 2026

ragesaq added 2 commits June 16, 2026 06:26

shakkernerd self-assigned this Jun 16, 2026

steipete added 6 commits July 1, 2026 08:15

feat: integrate durable agent activity

8820892

build: refresh embedded web assets

f40251a

fix: preserve agent activity invariants

9623b29

fix: keep message kind optional in SDK

8f4fece

fix: isolate agent activity by author

a35aa93

test: isolate agent activity browser coverage

af57acf

steipete changed the title ~~feat(web): cohesive agent-activity preamble + composer runtime controls + self-message bubbles~~ feat: add durable agent activity preambles Jul 1, 2026

steipete merged commit 933f6d7 into openclaw:main Jul 1, 2026
4 checks passed

ragesaq deleted the pr/agent-activity-cohesion branch July 1, 2026 08:36

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: add durable agent activity preambles#27

feat: add durable agent activity preambles#27
steipete merged 25 commits into
openclaw:mainfrom
ragesaq:pr/agent-activity-cohesion

ragesaq commented Jun 15, 2026 •

edited by steipete

Loading

Uh oh!

ragesaq commented Jun 15, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

ragesaq commented Jun 16, 2026

Uh oh!

clawsweeper Bot commented Jun 16, 2026 •

edited

Loading

Uh oh!

ragesaq commented Jun 16, 2026

Uh oh!

ragesaq commented Jun 16, 2026

Uh oh!

clawsweeper Bot commented Jun 16, 2026 •

edited

Loading

Uh oh!

steipete commented Jul 1, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

ragesaq commented Jun 15, 2026 • edited by steipete Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Security and invariants

Verification

Uh oh!

ragesaq commented Jun 15, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

ragesaq commented Jun 16, 2026

Uh oh!

clawsweeper Bot commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ragesaq commented Jun 16, 2026

Uh oh!

ragesaq commented Jun 16, 2026

Uh oh!

clawsweeper Bot commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

steipete commented Jul 1, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

ragesaq commented Jun 15, 2026 •

edited by steipete

Loading

clawsweeper Bot commented Jun 16, 2026 •

edited

Loading

clawsweeper Bot commented Jun 16, 2026 •

edited

Loading