Mission Control dashboard links 404: /copilot/tasks/<uuid> path doesn't exist; sessions actually live at /agents/tasks/<uuid>

[dai]

Describe the bug

The github.com Mission Control "Created by me" dashboard shows links to remote sessions that point to a non-existent URL path. Clicking the link returns a 404, but the session itself is alive and reachable from the CLI via copilot --resume=.

The dashboard renders each session as:

https://github.com/copilot/tasks/?q=is%3Aopen

but the actual cloud API is served at:

/agents/tasks/

The two paths are not equivalent. Web routing for /copilot/tasks/ returns 404 even though the task exists in the cloud backend.

Reproduction

1. Open https://github.com/copilot -> "Created by me" (or any session list).

2. Click the first remote session in the list.

3. Observe: browser shows a 404 / page-not-found for the task.

4. In a terminal on the same account, run:

   copilot --resume=05db5cb6-fea4-43ee-8914-4460ca096ca3

5. Observe: CLI prints "Remote control connected" and joins the same session, MCP servers reload successfully.

API evidence (collected 2026-09-09)

Endpoint	Auth	Status
GET /copilot/tasks/05db5cb6-fea4-43ee-8914-4460ca096ca3	no	404
GET /agents/tasks/05db5cb6-fea4-43ee-8914-4460ca096ca3	no	400 (bad request)
GET /agents/tasks/05db5cb6-fea4-43ee-8914-4460ca096ca3	yes	200
GET https://github.com/copilot/tasks/05db5cb6-...	n/a	404 (web route)

The CLI itself prints https://github.com/copilot/tasks/<uuid> after a successful resume, which suggests the URL template is generated client-side from a slug that omits the /agents/ prefix, instead of being derived from the actual API endpoint the CLI talks to.

Expected behavior

Clicking a session link in the Mission Control dashboard should land on a working detail page that resolves to the same task the CLI sees.

Affected version

• GitHub Copilot CLI: 1.0.84-3 (prerelease, 2026-09-09)
• Browser: latest Chrome on Windows
• github.com web app: current

Additional context

• The dashboard lists 320 active + 292 completed sessions on this account, with many entries under "No repository" -- likely accumulated test/orphan tasks from prior CLI versions and /remote experiments.
• The CLI's --resume\ works fine; only the web frontend's URL composition is broken.
• Possibly related to issue /remote command generates broken session URL (404) for enterprise users #2813 (URL format regression), but /remote command generates broken session URL (404) for enterprise users #2813 was fixed in v1.0.33 and the URL shape here is post-fix; this is a separate routing failure rather than a URL-shape failure.

[dai]

Follow-up: same task was filed twice within 5 seconds

While preparing the reproduction for this report, the same body was posted twice under my account (dai) within a 5-second window:

Issue	Created (UTC)	Markdown state
#4774	2026-09-09T03:55:21Z	backslash-escaped triple backticks (\\\\)
#4775	2026-09-09T03:55:26Z	clean triple backticks

• Both posted via the same gh token, neither triggered explicitly by a human typing.
• Mission Control dashboard links 404: /copilot/tasks/<uuid> path doesn't exist; sessions actually live at /agents/tasks/<uuid> #4774 was closed by the user (stateReason: COMPLETED) after the duplicate was spotted.
• No scheduled workflows exist (list_workflows returned []).
• The only plausible cause is a second agent / parallel process firing gh issue create without idempotency.

This pattern matches the dashboard symptom in the original report: the user has been observing that "the same session is always generated twice" and that "this session's output appears twice". Both observations point at the same root cause:

An idempotency / dedup gap in whatever path produces and dispatches Copilot task artifacts — write-side (issue filing) and read-side (dashboard listing) are two faces of the same bug.

If #4774 / #4775 are the only duplicate issue filings you can find under dai, this is more evidence the bug is specifically in the Mission Control / remote-session subsystem rather than generic gh retry behaviour.

Suggested fix surface (for triage)

• The endpoint that returns /agents/tasks/ for the CLI side and the dashboard's /copilot/tasks/ link generator may not share a single source of truth for "what sessions exist".
• Add an idempotency key (e.g., the task UUID) to whatever event dispatches these write actions so a double-fire collapses to one record.
• The dashboard should validate that any task UUID it renders as a link actually resolves via the API before showing the row, instead of caching a stale entry.

Happy to provide additional logs or test cases if useful.

[dai]

Cross-reference: root cause is in copilot-sdk

This issue is a downstream symptom of copilot-sdk#2560 ("CopilotClient.start() is not single-flight — concurrent auto-start spawns duplicate CLI processes"), which was opened 2026-09-08 and is already being fixed in PRs #2561 and #2570.

\
+-------------------------------+
| copilot-sdk#2560 |
| CopilotClient.start() lacks |
| single-flight guard |
+---------------+---------------+
|
+-----------------------+-----------------------+
| |
v v
+-----------------------+ +---------------------------+
| This issue (copilot- | | Duplicate issue creation |
| cli#4775): dashboard | | (copilot-cli#4774 was |
| 404 for valid task | | closed by reporter as |
| URLs | | COMPLETED) |
+-----------------------+ +---------------------------+
\\

When \CopilotClient.start()\ runs concurrently, two CLI processes are spawned. Each one independently:

1. Registers a session with the cloud via /agents/tasks/\ — so the cloud has two records.
2. Generates a URL of the form \https://github.com/copilot/tasks/\ — but only one survives as a working link.
3. Can independently fire \gh issue create\ — which is how this very issue ended up filed as Mission Control dashboard links 404: /copilot/tasks/<uuid> path doesn't exist; sessions actually live at /agents/tasks/<uuid> #4774 and Mission Control dashboard links 404: /copilot/tasks/<uuid> path doesn't exist; sessions actually live at /agents/tasks/<uuid> #4775 within 5 seconds under the same user account.

The user-visible observations all map cleanly to the same SDK-level bug:

• "Same session always generates twice" → dual \createSession()\ → dual CLI processes.
• "5-second gap between two identical issues" → each CLI independently posted.
• "SDK and CLI versions differ" → two CLI binaries racing the same client.

Fixing #2560 should make both #4775 and the duplicate-firing symptom disappear simultaneously.

cc: anyone who can shepherd #2561 / #2570 to merge.

[Tak2121]

سلام ببین من خودم هستم گوه خوردم که یک کیف پول هوشمند گرفتم الان ۵ ماه که اسیرم تا بخوام از توکن های خودم استفاده کنم تورو خدا کمکم کنید بخدا دیروز رفتم گوشیمو فروختم واسه این اکه خودم ارز دیجیتال دارم ولی نمیتونم استفاده کنم و دیگران به راحتی دارن از سرمایه من دزدی میکنند در تاریخ چهارشنبه ۹ سپتامبر ۲۰۲۶،‏ ۰۷:۳۹ dai ***@***.***> نوشت:

…

*dai* left a comment (github/copilot-cli#4775) <#4775 (comment)> Follow-up: same task was filed twice within 5 seconds While preparing the reproduction for this report, the same body was posted *twice* under my account (dai) within a 5-second window: Issue Created (UTC) Markdown state #4774 <#4774> 2026-09-09T03:55:21Z backslash-escaped triple backticks (\\\\) #4775 <#4775> 2026-09-09T03:55:26Z clean triple backticks - Both posted via the same gh token, neither triggered explicitly by a human typing. - #4774 <#4774> was closed by the user (stateReason: COMPLETED) after the duplicate was spotted. - No scheduled workflows exist (list_workflows returned []). - The only plausible cause is a *second agent / parallel process firing gh issue create without idempotency*. This pattern matches the dashboard symptom in the original report: the user has been observing that *"the same session is always generated twice"* and that *"this session's output appears twice"*. Both observations point at the same root cause: *An idempotency / dedup gap in whatever path produces and dispatches Copilot task artifacts — write-side (issue filing) and read-side (dashboard listing) are two faces of the same bug.* If #4774 <#4774> / #4775 <#4775> are the only duplicate issue filings you can find under dai, this is more evidence the bug is specifically in the Mission Control / remote-session subsystem rather than generic gh retry behaviour. Suggested fix surface (for triage) - The endpoint that returns /agents/tasks/ for the CLI side and the dashboard's /copilot/tasks/ link generator may not share a single source of truth for "what sessions exist". - Add an idempotency key (e.g., the task UUID) to whatever event dispatches these write actions so a double-fire collapses to one record. - The dashboard should validate that any task UUID it renders as a link actually resolves via the API before showing the row, instead of caching a stale entry. Happy to provide additional logs or test cases if useful. — Reply to this email directly, view it on GitHub <#4775?email_source=notifications&email_token=BMBZYCW7LVE5D5FH66YLPBD5ODJWRA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKNJZGU3DGNZYGIY2M4TFMFZW63VKON2WE43DOJUWEZLEUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#issuecomment-5595637821>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BMBZYCUUUNHZ6B7WZMUF6VL5ODJWRAVCNFSNUABFKJSXA33TNF2G64TZHM2TQNJYGYYDMNRUHNEXG43VMU5TKMZZGQ3TGMRYGIZ2C5QC> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>