Repository navigation
Conversation
A crew that runs unattended accumulates the state an operator actually wants:
how many workers it is holding, which one is stuck, what it is waiting on. The
drawer showed activity counts, a path list and config instead, none of which
answer those. This adds a free-form HTML webview that the crew fills in at
runtime.
The split that makes free-form HTML safe: a human authors the TEMPLATE --
versioned in the repository, or dropped on disk by the operator -- and the crew
publishes only DATA. The template directory is fenced from agent file tools, so
a crew cannot author its own markup; the only way it influences its webview is
by publishing a data object. The data lands as an inert application/json island,
escaped so a published string cannot close the script element early, and a
template reads it with JSON.parse and renders through textContent only. The
composed document is served behind a single-use minted URL into an iframe
sandboxed to allow-scripts and nothing else.
A crew whose name matches an installed template gets that template; a crew with
no template of its own falls back to a generic one that renders any data object.
That is the entire wiring mechanism -- installing <crew-name>.html is the act of
wiring a bespoke view up, so there is no registry, no mapping table and no
per-crew config to keep in sync.
The MCP server ships registered but gated off for every agent (opt_in), so a
default session pays nothing for the capability: not a tool-list entry, not a
line of context, not a spawned backend. An agent that should publish a panel is
granted the set in its own spec, and a refresh keeps an existing grant current
without ever minting one.
The publishing crew is derived from the vetted calling session and never from the
request body, so one crew cannot publish a webview that presents as another's. A
body field naming a crew is rejected by the schema rather than ignored, so a
caller is never left believing it retargeted the write. The drawer's read at GET
/api/members/{slug}/panel is cookie-authed; the /api/agent-panel writes sit
behind the MCP-only prefix.
Published field order is preserved end to end. Field order is presentation -- a
template renders a stat strip in key order -- so the record is dumped without
sorting keys instead of silently alphabetising the operator's dashboard.
# Conflicts: # docs/feature-map/README.md # src/kiro_crew/sandbox.py # src/kiro_crew/security/paths.py # website/src/i18n/locales/bn.json # website/src/i18n/locales/de.json # website/src/i18n/locales/es.json # website/src/i18n/locales/fr.json # website/src/i18n/locales/hi.json # website/src/i18n/locales/it.json # website/src/i18n/locales/ja.json # website/src/i18n/locales/ko.json # website/src/i18n/locales/pt.json # website/src/i18n/locales/ru.json # website/src/i18n/locales/zh-CN.json
|
Intent: Let an owner create tasks and steer a Crew Member in the existing member chat, and monitor a published task snapshot through the Tasks template in Crew summary's dashboard. Connect existing scoped worker dispatch to the real ledger while preserving private memory and ordinary tool approval. Not a goal: A separate task database, native task action controls, archive storage, or a continuously polled board. User-directed scope change: after comparing this feature with #8468, the owner explicitly selected "Template + existing chat (recommended)" for the first release. That choice replaces the earlier native Tasks tab and archive intent. This PR builds on #8468 and must be retargeted to main after that framework lands. |
Design Review (Fable 5) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS Sound gateway-sourced snapshot and verified-binding design; the new tool-exposure preflight is a parallel resolver that can silently drift from real backend semantics. Watch
Suggestions
[DESIGN-REVIEWED] 3e518c0 |
UX Review (Fable 5) — 🟡 CONCERNSUX-level review of All evidence is in hand. The change is a read-only "Tasks" kanban snapshot template rendered in the existing sandboxed member webview, with all strings localized across 13 locales; the blind read never ran because no screenshots were committed to the repository (the PR's captures live only as description attachments). UX-Verdict: CONCERNS Copy and states are solid, but zero screenshots were committed, so the blind read never ran — no cold-read evidence for the entire new Tasks board. Watch
Evidence gaps
Suggestions
[UX-REVIEWED] 3e518c0 |
First Principles Review (Fable 5) — 🟡 CONCERNSPremise-level review of First-Principles-Verdict: CONCERNS The member's dispatch loop moves to Not justified as shipped
What this change shipsInventory (10 of >10 items) — 7 justifiedIntent: let an owner create tasks in the member's chat and watch their progress on a board. ADDITION.
Not listed: Watch
Subtractions
[FIRST-PRINCIPLES-REVIEWED] 3e518c0 |
Opus 4.8 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
d057b62 to
dbe402b
Compare
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
558401f to
72ef806
Compare
|
Maintainer note: PR #11780 (merged 2026-09-19, Already conflicting; when rebasing please also update |
4db3d57 to
8af150d
Compare
|
🤖 Kiro Crew [operator: dwu96#8c08bee0]: This PR has been inactive for 7+ days. I reviewed the blockers but they require your input:
For the record, nothing else is blocking: the security scan is clean, and all seven review lanes (GPT 5.6, Opus 4.8, First Principles, Design, UX, Automated Rule Check, SAST) pass. When you've addressed these, the pipeline will re-assess on its next cycle. Add |
|
Triage note (automated PR-overlap scan, 2026-10-05). Checked against Status: rebase needed, and keep developing. This is not a duplicate, and main does not already cover it.
Overlap to discuss before both land: #16603 ("a project report beside every crewmate's chat") also renders the work ledger in the crewmate Dashboard tab. It adds a built-in This scan found no breaking change and no P0 in this PR's own commit. |
Problem / Motivation
Crew Members need a clear way to show task progress while owners create work and give instructions in the member’s existing chat.
Depends on #8468. This PR targets
feat/crew-webviewfor review. Merge it only after the dashboard framework lands and this PR is retargeted tomain; full CI runs only for PRs targetingmain.Why it matters
Owners should see the same task facts that their member uses. A worker saying it is done must not look like an accepted result, and failed dispatch preparation must leave the task available to retry.
What changed (motivation → approach → change)
The bundled Tasks template shows a published ledger snapshot in Crew summary → Dashboard. Owners create tasks, steer work and give acceptance decisions in the existing member chat. To do, In progress, Blocked, Review, Done and Closed come from ledger fields. Worker completion reaches Review; acceptance reaches Done. Cards show reports, criteria, decisions, verification results and evidence. Labels and dates are translated, and the document renders text inside the existing sandbox.
spawn_run(work_item_id=...)connects one task to a worker in the caller’s own ledger. The gateway prepares the worker and binds its generated identity before the first stream. Scoped dispatch defaults tokirocrew-workerand checks both reporting tools using the allocated provider’s actual backend and directory. Explicit tool exclusions override a broad allowlist. KAS uses its user-level spec and wildcard projection; the other supported backends use their project-aware resolver. The worker gets its own runtime identity, ordinary approval and inherited private memory. Briefs and reports require the binding and item to name the same worker; an interrupted bind cannot authorize an obsolete worker. Failed preparation leaves the task retryable. Member DMs receive work and panel tools without new approval grants. Their tracked-task procedure uses scoped spawn and a bounded patrol to check reports, steer stalls and publish updates.panel_publish(template="tasks", data={summary?})reads the verified caller’s own ledger. Only the optional summary comes from model-supplied data. The snapshot omits event history and worker session keys; an assignment flag is enough for its lanes. Publication refreshes the existing panel query. The document shows its publication time, missing or truncated data, and the chat path for changes or updates. The shared publication scrubber retains field-name context before persistence, removing named credentials while preserving surrounding prose and structure. This template replaces the proposed native member task table for the first release.🟩 added · 🟨 changed · 🟥 removed · 🟦 unchanged
Owners use the member chat to direct work and the published board to see its recorded progress.
Tests
main.Manual verification
The real Members page, template, themes and i18n code were exercised with intercepted API responses. Browser verification covered the board, card evidence, sandbox grants, instructions sent through member chat, dark and light themes, task content at 390px and 320px without horizontal overflow, and empty, unavailable and truncated snapshots. The expanded card shows distinct Pull request, Evidence and Verification result labels; the board names chat as the place to accept results. The recording demonstrates this fixture-based flow. The screenshots were visually inspected; this is not a human usability study. The binding correction changes no UI code, so the attached captures still show the current interface.
Screenshots / video
Chat instructions, light theme, narrow layout and empty state
task-template-flow.webm
Related Issues
No linked issue: requested directly to create tasks, steer the member and monitor progress, with the first release using a template and existing chat.
Checklist
main— pending feat(members): give every crew a configurable webview in its drawer #8468 and retargeting; local full-backend limitations are listed above.