Skip to content

fix(spawn): admit the host operator's cli spawn without a session - #15593

Closed
koenvincent wants to merge 1 commit into
kirodotdev:mainfrom
koenvincent:fix/cli-spawn-identityless-scope
Closed

koenvincent wants to merge 1 commit into
kirodotdev:mainfrom
koenvincent:fix/cli-spawn-identityless-scope

Conversation

@koenvincent

@koenvincent koenvincent commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: kirocrew spawn run starts a subagent again instead of always failing.

On a 0.8.0rc1 gateway, every kirocrew spawn run '<task>' fails with 409 member_identity_unavailable ("The execution identity is unavailable; Global memory was not used."). No run is created.

Why it matters

On a headless gateway (no dashboard attached) the CLI is the only local way to start a subagent, so there is no way to start one at all. Scripts and SSM-driven automation that call kirocrew spawn run are hit every time, with an error that points at memory identity rather than at the CLI.

Not a goal

  • No change to how header-less internal callers are treated on any other route (directives, keepalive, tool policy, workflows, spawn continue/batch).
  • No new CLI flag for passing a session key.
  • No change to spawn approval: a CLI spawn faces the ordinary spawn prompt, so a headless gateway still needs auto_approve_subagent_spawn or a dashboard to approve it. Whether the owner's terminal may skip the prompt is #16313.

What changed (motivation → approach → change)

  • Symptom → cause. The CLI sent the internal secret with no session and no parent. api_spawn claimed the body default "", the verified scope had session=None, and internal_memory_scope compared "" == None, so every CLI spawn got the 409.
  • Approach. The CLI is the dashboard owner at a terminal, so it now presents the owner's own credential instead of asking the gateway to infer an identity from a session-less internal request. The shared identity check (internal_memory_scope) is unchanged, so directives, keepalive, tool policy and workflows keep their behaviour. Normalising "" to None there was rejected: it would widen exactly those routes.
Piece Change
Who the CLI is kirocrew spawn authenticates as the dashboard owner (new owner_gateway_client), not as an anonymous internal-secret caller
How it gets there a 2-minute owner token from GET /api/token/local?origin=cli, over the dashboard unix socket only (no TCP path), after the CLI checks that the socket's peer is the gateway recorded for the port (run/gateway-<port>.pid and its start identity)
What marks a CLI run the token's signed origin: "cli" claim (read with the new token_auth.validated_token_origin), for the owner principal only, recorded on the run's ExecutionContext.origin; a CLI spawn that names a parent_session is 400 invalid_origin
What the CLI token is minted with no one-time-link nonce, so it cannot be opened as a link and a burst of CLI calls cannot evict a pending login or channel link
How a CLI run is governed as surface cli (cli_chat, subagent.spawn_governance_key) on every agent name it answers to (subagent.spawn_policy_agents), at admission and at the run-time re-check; a spawn that names no agent is checked as kirocrew and runs kirocrew by name (run.py claims it explicitly), so the warm pool's session.pool_agent / agent.default_agent cannot run in its place
Its spawn prompt the ordinary one, as for any other caller: with no dashboard open and no auto_approve_subagent_spawn or Trust it waits for approval or is refused, and the poll says so. Skipping it for the owner's terminal is a separate decision (#16313, RFC then opt-in)
What does not inherit it a run the CLI run's model spawns (derive_execution clears origin) and a schedule it creates (bind_cron_memory)
Parentless runs the owner principal's: an internal caller with no session can neither start one (409 from the shared identity check, as on main) nor control one (_run_belongs_to_caller, list included)
The poll a 5xx (the gateway's taskq_unavailable "retry shortly") or a poll whose answer was lost is asked again up to _SPAWN_POLL_RETRIES times, after Retry-After; any other failure ends it with the gateway's or the transport's error and says the run may still be going; exits nonzero for any ending but completed (a stopped run, and a run whose ending nothing recorded, are not a success); hints at spawn and tool prompts
kirocrew spawn list the queued rows #16012 added (?queued=1), over the owner transport, every field confined to its line
Gateway pieces the CLI path needs /api/token/local registers with the routes both servers share, so a headless --slack-only gateway serves it; both servers name the unix socket for the bound port, so --port auto works; GET /api/spawn/{id} done answers carry outcome, and both read paths report awaiting_tool_approval
  • Parentless run controls. The Not-a-goal bullet names the shared identity check on other routes; that check is unchanged. The spawn routes' own ownership rule does change: on main an identity-less internal caller owned every parentless run because the old CLI created them. Parentless runs are now created by the owner principal, so leaving that arm would let any identity-less internal caller act on the owner's runs. It is narrowed in this PR rather than later because creation and control move together.

Backwards compatibility

Breaking:

  • An internal-secret caller with no X-Session-Key can no longer start, list, read, steer, release, retry or delete a parentless run. Writer sweep on origin/main @ 3d465a6: the only such callers were the old kirocrew spawn (replaced here), an MCP stdio server that lost its session (now refused with the existing identity-gap message) and code_review_sage's _probe, which only checks that a gateway answers.
  • kirocrew spawn is POSIX-only for now. kirocrew spawn run/list has no TCP path. Where the gateway serves no dashboard socket (Windows, or a socket bind that failed, for example a data-home path too long for sun_path), the CLI says so and points at the dashboard. spawn list worked over TCP on main; an owner transport for socketless gateways is #16314. cli.md and subagent.md state it.
  • New refusal, every surface: a spawn that names no agent, no member and whose parent records no template is checked against capabilities.spawn.scopes.agents as kirocrew at admission and at the run-time re-check, and runs kirocrew by name. On main it was checked at neither, and it ran whatever the session claim picked for no agent (the warm pool's session.pool_agent or agent.default_agent on a pool hit, kirocrew on a cold start). So a profile whose capabilities.spawn.scopes.agents is an allow-list without kirocrew refuses it at admission (agent 'kirocrew' not permitted by spawn policy); remedy: add kirocrew to the allow-list. The agent that answers changes too: an operator who set agent.default_agent (or session.pool_agent) no longer gets that agent for a nameless subagent spawn on a warm-pool hit; it runs kirocrew and starts cold, because the agent governance checked must be the agent that runs. Remedy: name the agent explicitly on the spawn (agent=). Pinned by test_a_spawn_that_names_no_agent_is_checked_as_the_default_agent and test_a_run_naming_no_agent_runs_the_agent_its_scope_checked; governance.md states it.
  • kirocrew spawn run exits 1 for a stopped run, which main reported as a success.
  • Under --port auto the dashboard socket is named for the bound port (dashboard-<port>.sock), not dashboard-0.sock.

Tests

  • test/test_spawn_cli_owner.py (new): the CLI mints an origin=cli owner token over the checked socket and never uses TCP (test_spawn_run_mints_a_cli_owner_token_over_the_checked_socket, test_a_gateway_with_no_socket_is_named_and_tcp_is_never_tried); each mint refusal names its remedy; the POST and mint timeouts say whether anything was sent; a 503 with Retry-After, a read timeout and a cut-off answer are asked again (test_the_poll_asks_again_after_a_transient_failure) and give up after the bounded retries (test_the_poll_gives_up_after_its_retries); a poll that never reached the gateway, a gone socket, a listener that is not the gateway, or a 4xx ends it and says the run may still be going (test_a_lost_poll_ends_and_says_the_run_may_still_be_going); a stopped run and an unrecorded ending exit 1 (test_a_stopped_run_is_not_reported_as_success, test_an_ending_nothing_recorded_is_not_reported_as_success); a 403 earns one fresh mint; a clock jump renews the token; a listener that is not the gateway receives no credential; only the CLI token marks a CLI run and it carries no link nonce; a CLI spawn naming a parent is refused; a session-less internal caller starts and controls no parentless run, and the shared identity check keeps its 409; a CLI spawn faces the ordinary spawn prompt (test_a_cli_spawn_faces_the_spawn_prompt); a child and a schedule do not inherit the origin; a surface:cli profile governs the CLI spawn, a retry of it and the run-time re-check; a spawn naming no agent is checked as kirocrew on any surface; a spawn that names its agent is checked again when its run starts, so a scope tightened while it waited for approval stops it (test_a_named_agent_is_rechecked_when_its_run_starts; the run-time re-check used to cover only inherited names); the warm pool matches on the agent that runs, so a pool filled for no agent serves a nameless spawn (test_the_default_agent_by_name_claims_a_pool_filled_for_no_agent, session_pool._claim_from_pool); both servers serve the mint. test_kirocrew_spawn_end_to_end runs the real CLI against the real middleware, owner bootstrap and peer check, through a clock jump and a logout, and sends the old internal-secret shape, which gets the 409.
  • test_dashboard_server_startup_coverage.py::TestStartDashboardWiring::test_backends_spawn_with_the_reserved_ports_evidence_exported and test_dashboard_server_coverage.py::TestStartApiServerResidualPaths::test_the_unix_socket_is_named_for_the_bound_port: both entrypoints name the unix socket for the bound port.
  • test_taskq_admission_integration.py::test_a_cli_origin_row_keeps_its_origin_and_faces_the_prompt.
  • Updated: test_member_memory_api.py (test_run_controls_fail_closed_for_a_caller_with_no_session_identity, test_spawn_list_shows_an_identity_less_caller_nothing), test_subagent_continuable.py (the inherited-template refusal lands at admission; a continuation of a default-template run claims kirocrew by name), test_subagent_spawn_approval_parked_6484.py, test_cli.py and test_cli_commands_coverage.py (owner-token flow, shared fake in test/spawn_cli_fakes.py).
  • The new guards were mutated locally and their tests went red.

Manual verification

Observed the 409 from kirocrew spawn run --async on a 0.8.0rc1 gateway. test_kirocrew_spawn_end_to_end drives the real CLI code against a real unix site; a live gateway run with this build is still to do.

Not built here: cli_server and pod/runtime_client keep their own mints; with agent.approval_mode: interactive and no dashboard open, a mid-run tool prompt waits silently up to 2 h apart from the CLI's hint; the dashboard Spawn button keeps its empty-key governance; no prompt waiver for the CLI (#16313); no CLI spawn path on Windows (#16314); with the sandbox off, or on Windows, there is no agent/operator boundary for this path, as elsewhere in those modes.

Related Issues

Part of #16480

No linked issue: the bug was reported on 0.8.0rc1. The earlier 0.8 backport, #15824, is merged.

Pattern harvest

Rule candidate: review-prompt
Pattern: a terminal command that acts for the owner should present the owner's credential, not ask the gateway to infer identity from the absence of one.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable): subagent.md § CLI states the rule; dashboard-token-auth.md documents the shared mint route and the origin=cli token; security.md, governance.md, taskq.md, learn-cron-dashboard.md and slack-gateway.md link to it; code-style.md lists the two new constants.py names; the kirocrew-commands skill notes the commands are the owner's
  • No secrets, credentials, or internal references in the diff

@koenvincent
koenvincent requested a review from a team as a code owner September 30, 2026 21:44
@dwu96

dwu96 commented Sep 30, 2026

Copy link
Copy Markdown
Contributor

👋 Hi! This PR's description is missing some required sections from our PR template. Workflow runs won't be auto-approved until the description is updated.

Missing sections:

  • ## Problem / Motivation
  • ## Why it matters
  • ## What changed
  • ## Tests

Please update your PR description to include these sections, then push or re-save the description. The workflows will be approved on the next cycle.

@github-actions github-actions Bot added fork Pull request from a fork (external contributor) readiness: checking Automated validation is still running labels Sep 30, 2026
@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review (fork) — ✅ no blocking findings

Reviewed 502e9add57f8d292ed9079e9f6c7499afcf2439b via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 502e9ad

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5, fork) — 🟡 CONCERNS

Design-level review of 502e9add57f8d292ed9079e9f6c7499afcf2439b via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

A CLI-auth fix also tightens spawn governance and changes which agent runs on every surface. That should be its own PR, and the fix still leaves Windows broken.

Watch

  • Governance change hidden inside a CLI fix. The PR says "New refusal, every surface". Through spawn_policy_agents and run.py's agent = agent or DEFAULT_AGENT_NAME, every nameless spawn_run from chat, MCP, Slack or a crew is now checked as kirocrew and runs as kirocrew. On a warm-pool hit it no longer gets agent.default_agent. So the chain is: a profile with an agents allow-list that lacks kirocrew → every model-initiated nameless spawn is refused. The CLI goal does not need this: only the CLI-origin path has to be checked under cli_chat. The session_pool._claim_from_pool rematch rides along with it.
    Clears when: the every-surface admission/run-time agent change and the pool match move to their own PR, or are limited here to origin == CLI_ORIGIN runs.
  • The goal is not met where there is no socket, and spawn list loses ground there. "kirocrew spawn is POSIX-only for now": on Windows spawn run stays broken, and spawn list, which worked over TCP on main, stops working. That is reversible and tracked in kirocrew spawn: owner transport for gateways with no dashboard socket (Windows) #16314, but the headline fix does not reach those platforms.
    Clears when: kirocrew spawn: owner transport for gateways with no dashboard socket (Windows) #16314 is linked as a follow-up that owns Windows parity, or an owner transport with a verified TCP peer lands before release.

Suggestions

  • The owner-token client is now the third socket client, after app_lifecycle_client and memory_forget_client. Folding app_lifecycle_client onto owner_gateway_client in the stated follow-up removes the copy that has no peer check.

[DESIGN-REVIEWED] 502e9ad

Description read

[DESCRIPTION-READ] f92df90c1f013d0a2d086f5f5472e9056ba28a953e2ae2feb480f44dda2a08cb

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5, fork) — 🔴 BLOCK

Premise-level review of 502e9add57f8d292ed9079e9f6c7499afcf2439b via the fork AI-review pipeline — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: BLOCK

product-shape change without accepted RFC: on every surface, a spawn that names no agent now runs kirocrew, not the configured pool/default agent. The Goal never asked for that.

Not justified as shipped

  • Item 6 — oversized: the Goal is "kirocrew spawn run starts a subagent again", yet the description says "an operator who set agent.default_agent (or session.pool_agent) no longer gets that agent for a nameless subagent spawn" (run.py agent = agent or DEFAULT_AGENT_NAME).
    Subtraction: drop the run.py default and the _claim_from_pool rewrite. Check governance against the agent the pool actually runs, or limit the kirocrew default to origin == CLI_ORIGIN.
    Clears when: the change applies only to CLI-origin runs, or a decision is recorded (see Blockers).
  • Item 9 — duplicate of src/kiro_crew/app_lifecycle_client.py:154 (1 sibling inline /api/token/local mint): "kirocrew app … mints the same way … folding it onto this module is a separate change."
    Subtraction: delete the inline mint in _post_lifecycle_action and call mint_owner_token.
    Clears when: one mint path remains.
  • Item 5 — rides along: tool-approval hint fields; not needed to start a run.
  • Item 7 — rides along: poll retries and a nonzero exit for stopped runs change script-visible behaviour.

What this change ships

Inventory (10 items) — 5 justified

Intent: let the host operator start a subagent from the terminal. FIX.

  1. kirocrew spawn run authenticates as the owner with an origin=cli token over the checked unix socket — justified
  2. Headless and --port auto gateways serve the mint and the socket — justified
  3. CLI runs are governed as surface cli, and spawned children or crons do not inherit that — justified
  4. Internal callers with no session no longer own parentless runs — justified
  5. Status and list report awaiting_tool_approval, and the poll prints hints — rides along
  6. Nameless spawns on every surface are checked as kirocrew and run kirocrew, skipping the warm pool's agent — oversized
  7. Poll retries 5xx and lost answers; a stopped run exits 1 — rides along
  8. spawn list loses TCP, so it is unavailable on Windows and socketless gateways — justified
  9. New owner_gateway_client mint beside the existing one — duplicate of app_lifecycle_client.py
  10. spawn list fields are confined to their line — justified

Blockers

product-shape change without accepted RFC: the default agent for nameless spawns. The description says "The agent that answers changes too". rfc-status.txt lists no RFC that covers spawn-default agent selection.
Clears when: an RFC covering the nameless-spawn agent default is merged non-draft on the base (then push or rebase), or a maintainer posts /ai-review override first-principles 502e9add57f8d292ed9079e9f6c7499afcf2439b: <decision>.

[FIRST-PRINCIPLES-REVIEWED] 502e9ad

Description read

[DESCRIPTION-READ] c1182493d0034a1c3af70740970d11150fa9c1ff9fa2f7a22adc91602947a468

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Sep 30, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review (fork) — ✅ no blocking findings

Reviewed 502e9add57f8d292ed9079e9f6c7499afcf2439b via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 502e9ad

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 1, 2026
@koenvincent
koenvincent force-pushed the fix/cli-spawn-identityless-scope branch from 82ca209 to f3a2a49 Compare October 1, 2026 05:53
@bolichen97

Copy link
Copy Markdown
Contributor

Thanks @koenvincent. The root-cause analysis ("" vs None in internal_memory_scope) is right, and your commit is kept as is. I pushed one follow-up commit on top, rebased on current main:

What it changes. The no-session exemption now needs the CLI's whole shape AND a host peer process (messaging._is_host_cli_spawn):

  • no X-Session-Key, no parent_session, no X-Internal-Caller: every MCP stdio server sets that header (mcp_core._caller_header), so a pooled server that lost its session is no longer read as the CLI. This is the Opus 5 BLOCK at messaging.py:663.
  • host provenance via member_memory_auth.local_owner_bootstrap_allowed (the check /api/token/local already uses). .local_secret is a VISIBLE sandbox leaf, so an agent's own shell can run kirocrew spawn run. /api/memory/recall refuses a session-less internal caller (400 missing_session_key), so without this check the CLI exemption would have been a new route from a crew-member or incognito session to a Global, persistent child.

Tests. A non-host peer, a session-unbound MCP caller, and a caller with a session key but no parent each stay 409 member_identity_unavailable, and the host CLI is admitted. I removed each clause in turn and the matching case went red, which also covers the GPT 5.6 mutation finding. subagent.md documents the predicate and the residual: with the sandbox off, an agent's shell is a host process and cannot be told apart from the CLI.

Not changed here, will track separately:

  • A related hardening of run-control scoping for session-less callers is tracked separately.
  • On a headless gateway with default hooks, an admitted parentless spawn still needs a spawn approval, so kirocrew spawn run is only fully usable there with auto_approve_subagent_spawn or a dashboard attached.
  • The CLI's 5 s POST timeout and its 404-as-"lost connection" poll message are older CLI UX issues.

A backport to release/0.8.0 follows in a separate PR, since 0.8.0rc1 has the same 409.

@bolichen97
bolichen97 force-pushed the fix/cli-spawn-identityless-scope branch from f3a2a49 to 581cbba Compare October 1, 2026 10:52
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Oct 1, 2026
@bolichen97
bolichen97 force-pushed the fix/cli-spawn-identityless-scope branch from 581cbba to 14cdba7 Compare October 2, 2026 16:40
@bolichen97 bolichen97 changed the title fix(spawn): admit kirocrew spawn run without a session key fix(spawn): admit the host operator's cli spawn without a session Oct 2, 2026
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: action required A blocking check or review needs attention labels Oct 2, 2026
@bolichen97

Copy link
Copy Markdown
Contributor

Pushed 14cdba7290, one commit (author @koenvincent, Co-authored-by Bolin Chen), rebased onto main at 07e6f0b345. It replaces the earlier two commits.

What this push covers:

  • Redesign. kirocrew spawn now authenticates as the dashboard owner. It uses a short-lived owner token from /api/token/local?origin=cli, sent over the dashboard unix socket only, and only after a peer check against the recorded gateway pid. The session-less internal-secret admission and its host-process predicate are gone. The shared identity check is unchanged. The PR body has been rewritten to describe the shipped design.
  • Review rounds. The internal review findings are applied: origin is not inherited by child runs or schedules; governance checks both the template and the resolved agent; the poll handles restarts, unseen runs, outcome-less answers and stopped runs; the prompt waiver is withheld from continuations and dropped when a queued run is cancelled; plus spec and test coverage.
  • Rebase. kirocrew spawn list now renders the queued rows from fix(subagent): show gate-deferred spawns as queued and hold synthesis #16012 (?queued=1) over the owner transport.
  • Cleanup. The CLI prompt waiver set is now a declared SubagentManager attribute.

Below are dispositions for the findings the lanes posted on 581cbba1.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=f4835c4b5f1c fixed in 14cdba7: app backends are misclassified as the host CLI

fixed: the cited predicate (_is_host_cli_spawn, which delegated to local_owner_bootstrap_allowed for an internal-secret caller) is deleted. No internal-secret caller without a session can start a parentless run any more (409 member_identity_unavailable, pinned in test_spawn_cli_owner.py).
The CLI now presents an owner token from the existing /api/token/local mint. That mint's admission rules are unchanged from main, so this PR adds no new holder of the owner credential.
origin=cli is a signed claim minted only there, and it is honoured only for the owner principal.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=2af0175e538b fixed in 14cdba7: admission no longer depends on X-Internal-Caller being absent

fixed: the redesign removes every admission decision that read the header. messaging.py no longer reads X-Internal-Caller for spawn admission.
The CLI authenticates with an owner token, not with the internal secret. So the attribution-only contract in mcp_shared.py is true as written and needs no change.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=5fff060a7d24 rebutted: the description is rewritten; a live gateway run is not possible in this environment

The description now matches the shipped design (owner token, peer check, origin claim), and the test names it cites exist in the diff.
The tests no longer monkeypatch the admission predicate. test_kirocrew_spawn_end_to_end runs the real CLI against the real middleware, owner bootstrap and kernel peer check over a real unix site, and the old internal-secret shape still gets the 409 there.
A run against a live built gateway is listed as still to do under Manual verification, for a maintainer with a deployment.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=e55839fe7b41 rebutted: the undeclared host-process gate is gone; the CLI's path is exercised end to end

The predicate this item names (no X-Internal-Caller plus local_owner_bootstrap_allowed peer provenance) is deleted in 14cdba7. The description states the replacement: an owner token over the checked unix socket.
test_kirocrew_spawn_end_to_end drives the real CLI code through the real middleware and peer check, so the motivating caller's path is executed, not mocked.
A 200 from a live deployment cannot be produced from this environment, and is listed as still to do under Manual verification.

@bolichen97

Copy link
Copy Markdown
Contributor

PID 1: docker/entrypoint.sh execs tini -- kirocrew gateway --no-open, so tini is PID 1 and the gateway is its child. The mint's PID-1 refusal reaches only a container started with no init, which the official image is not. In subagent.md, "as the official image runs it" qualifies the init (tini), not the refusal. Same ruling as 6c05a2388ae8, 402d32145ecd and 5582d559edc5.
Approval: the description's "Not a goal" says a CLI spawn faces the ordinary spawn prompt, so a headless gateway still needs auto_approve_subagent_spawn or a dashboard to approve it. Skipping the prompt for the owner's terminal is #16313. test_a_cli_spawn_faces_the_spawn_prompt pins that the prompt is not bypassed. What this PR fixes is the unconditional 409 that came before any of that.
Windows: 5964899894 accepts that kirocrew spawn list is socket-only and not available on Windows for now, with #16314 as the follow-up. Carried from span 42776df4749f.
This head (37e4d64) is a maintainer rebase of dc38783 onto main 51371a0. It resolves two conflicts, the MCP route-table pin (244 rows: main 242 plus GET/HEAD /api/token/local) and the cron/spawn rows in cli.md. No code changed, so the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=ca87ceff016f rebutted: the description is complete on the PR, and no reader dials dashboard-0.sock

The PR body is complete (about 12.5k characters, running through Tests, Manual verification, Related Issues and Checklist). The truncation is in this lane's capture, not in the description.
On the open question: under --port auto both entrypoints now resolve the bound port once and pass it to _start_unix_site (server.py: _start_unix_site(runner, _bound_port)), so the socket is dashboard-<bound>.sock. Every reader derives the path through dashboard_socket_path(port) from the port it resolves (mcp_core._socket_path_for, cli_server, app_lifecycle_client, owner_gateway_client). Nothing composes dashboard-0.sock. Pinned by test_the_unix_socket_is_named_for_the_bound_port and test_backends_spawn_with_the_reserved_ports_evidence_exported. The session-less internal-caller sweep is pinned by test_run_controls_fail_closed_for_a_caller_with_no_session_identity and test_spawn_list_shows_an_identity_less_caller_nothing.
This head (37e4d64) is a maintainer rebase of dc38783 onto main 51371a0. It resolves two conflicts, the MCP route-table pin (244 rows: main 242 plus GET/HEAD /api/token/local) and the cron/spawn rows in cli.md. No code changed, so the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=8d86a6682361 rebutted: maintainer ruling, a nameless spawn runs kirocrew on every surface

This item clears when "a maintainer records the all-surface change". That record exists: acceptance 5964899894, plus rulings 5964017525, 5977964351, 5988267324, 6007795939, 6038349655 and 6048628725. A spawn that names no agent is checked as kirocrew at admission and at the run-time re-check, and runs kirocrew by name, so the agent the ceiling checked is the agent that runs. On main a nameless spawn was checked at neither point, so a capabilities.spawn.scopes.agents allow-list did not bind it. Limiting the pin to origin == cli would leave that gap open on every other surface. Operators who relied on agent.default_agent substituting name the agent; Breaking states that remedy.
This head (37e4d64) is a maintainer rebase of dc38783 onto main 51371a0. It resolves two conflicts, the MCP route-table pin (244 rows: main 242 plus GET/HEAD /api/token/local) and the cron/spawn rows in cli.md. No code changed, so the finding is unchanged. Carried from span f284dd52dda7.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=189c919dfa0b rebutted: the poll retry, outcome exit and tool-approval hint are the restored blocking command's contract

On main, kirocrew spawn run exited 0 for a stopped run, so a script chaining on it read a false success. Breaking declares the nonzero exit, pinned by test_a_stopped_run_is_not_reported_as_success. The bounded retry (2 attempts) covers only answers the gateway labels retryable: a 5xx with Retry-After, or a GET whose answer was lost. Without it, one such answer ends a blocking wait while the run goes on, which invites a duplicate spawn (test_the_poll_asks_again_after_a_transient_failure, test_the_poll_gives_up_after_its_retries). awaiting_tool_approval is read from the gateway's existing pending-approval state, so a run parked on a tool prompt does not sit on "waiting for result..." with nothing for the user to act on (test_a_run_parked_on_a_tool_prompt_says_so_on_both_read_paths). These are the same rulings as spans 52086652dabf, d3a4ac92b30f and 654fba0fc5c2 on dc38783, and 6048628725 covers them.
This head (37e4d64) is a maintainer rebase of dc38783 onto main 51371a0. It resolves two conflicts, the MCP route-table pin (244 rows: main 242 plus GET/HEAD /api/token/local) and the cron/spawn rows in cli.md. No code changed, so the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor

Replacing toggle_app's inline mint with mint_owner_token changes kirocrew app, which is outside this PR's goal. The maintainer deferred that fold to #16350 (rulings 5964017525, 5977964351, 5988267324 and 6048628725), and the owner_gateway_client module docstring says so. The new client also pins the peer to the host's own run/gateway-<port>.pid record via run_marker before it sends the local secret. The inline app mint does not, so the fold is a behaviour change to kirocrew app and needs its own review.
This head (37e4d64) is a maintainer rebase of dc38783 onto main 51371a0. It resolves two conflicts, the MCP route-table pin (244 rows: main 242 plus GET/HEAD /api/token/local) and the cron/spawn rows in cli.md. No code changed, so the finding is unchanged. Carried from span a47f87d5931d.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=01601ed122ca rebutted: maintainer ruling, a nameless spawn runs kirocrew on every surface rather than agent.default_agent

The maintainer accepted this exact consequence on this PR: 5964899894 says "operators who relied on agent.default_agent or session.pool_agent substituting must name the agent, and the release notes carry that line from the Breaking section". Rulings 5964017525, 5977964351, 5988267324, 6007795939, 6038349655 and 6048628725 restate it. Breaking reports it to the operator.
Resolving the default from config instead would check one name at admission while get_or_create could still resolve another, through session.pool_agent on a pool hit or agent.default_agent on a cold one. The PR closes that check-one-run-another gap by pinning the name. The cold-start cost is the warm-pool half of this span, fixed on 634caf5: _claim_from_pool matches on the agent that runs, so a pool filled for no agent serves the nameless spawn. Pinned by test_the_default_agent_by_name_claims_a_pool_filled_for_no_agent.
This head (37e4d64) is a maintainer rebase of dc38783 onto main 51371a0. It resolves two conflicts, the MCP route-table pin (244 rows: main 242 plus GET/HEAD /api/token/local) and the cron/spawn rows in cli.md. No code changed, so the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor

Maintainer rebase pushed: 37e4d64 -> ddd2cd7, onto main aee834c. It applied cleanly with no conflicts, so no code changed and the route-table pin did not need a new digest. It is still one commit, authored by Koen Vincent, with the Co-authored-by trailer.

Why: on 37e4d64 the only reds were test_crew_recency (Backend Tests 3.12/1, macOS 1, Windows 1, and the Coverage jobs that follow from them). Those were red on main too, and #18021 has since fixed them on main.

Checked locally on the new head: black, isort, flake8 and mypy are clean. The targeted suites pass (test_crew_recency, the dashboard composition contract, test_spawn_cli_owner, the CLI, governance and subagent suites: 2866 passed). scripts/local-gate.py passed apart from test_build_desktop_managed_marker, which only times out on this host when HOME is pointed at a scratch directory (the mise node shim hangs there). That file passes 18/18 with the real HOME, and this PR does not touch it.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=309d4421612a rebutted: covering chat's spawn_run as well is the accepted point of the change, not a leak

On main a nameless spawn was checked at neither admission nor the run-time re-check, on any surface. So a capabilities.spawn.scopes.agents ceiling did not bind chat's spawn_run either. Limiting the kirocrew pin to the CLI would leave that gap open everywhere else. Maintainer acceptance 5964899894 names "every surface" explicitly.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=436a4a1ae8fe rebutted: a maintainer explicitly accepted the nameless-agent change in this PR's scope

The acceptance is recorded on this PR in 5964899894 ("A spawn that names no agent runs kirocrew on every surface"), and restated in rulings 5964017525, 5977964351, 5988267324, 6007795939, 6038349655 and 6048628725. The "New refusal, every surface" row in Backwards compatibility is that decision written down for operators, not an unflagged side effect. Carried from span 87eb819422ac.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=7cb1d44eae93 rebutted: the CLI fix needs the checked agent to be the agent that runs, and a maintainer accepted the bundle

kirocrew spawn run usually names no agent. Governing it as surface cli against a scopes.agents ceiling means checking some agent name. Unless that name is pinned, the run can be checked as one agent and run as another. So the pin is part of making the CLI governance honest, not a separate decision riding along. 5964899894 accepts the two together in this PR's scope, and rulings 5964017525 through 6048628725 restate it.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor

The suggestion is to track folding the two owner-mint clients as a named follow-up. That issue exists and is open: #16350, "Fold app_lifecycle_client's owner-token mint onto owner_gateway_client". It is also the deferral the maintainer recorded in 5964017525 and 6048628725. It is kept out of this PR because folding would add the run_marker peer check to kirocrew app, which is a behaviour change to a command this PR does not otherwise touch.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=ab7758fed7d2 rebutted: the new refusal is the allow-list finally binding, and Breaking names the remedy

Before this change, an allow-list that left out kirocrew did not stop a nameless spawn, because nothing checked it. Now it does, which is what the operator's ceiling says. Breaking gives the remedy: name the agent, or add kirocrew to the allow-list. A maintainer accepted the trade in 5964899894.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=cd7cca593a3e rebutted: the lost default-agent substitution is accepted, and the warm pool still serves the default case

5964899894 accepts it: "operators who relied on agent.default_agent or session.pool_agent substituting must name the agent". Breaking carries that line into the release notes. Resolving the default from config instead would check one name at admission while get_or_create could still run another (session.pool_agent on a pool hit, agent.default_agent on a cold start).
The warm pool: _claim_from_pool (session_pool.py) compares both sides as the agent that runs, so a pool filled for no agent still serves a nameless spawn. That is pinned by test_the_default_agent_by_name_claims_a_pool_filled_for_no_agent. Only a pool deliberately filled for a different agent starts cold, and that is the accepted case.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=e3b068318850 rebutted: this item's own clear condition is met, because a maintainer explicitly accepted the change here

The item clears when "a maintainer explicitly accepts them as in scope here". That acceptance is 5964899894, which covers the spawn_policy_agents default, the run.py default and the _claim_from_pool match. It is restated in rulings 5964017525, 5977964351, 5988267324, 6007795939, 6038349655 and 6048628725.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor

Replacing toggle_app's inline mint with mint_owner_token changes kirocrew app, which is outside this PR's goal. The maintainer deferred that fold to #16350 (rulings 5964017525, 5977964351, 5988267324 and 6048628725), and the owner_gateway_client module docstring says so. The two flows also differ on trust. Before it sends the local secret, the new client checks that the peer is the host's own gateway, using the run/gateway-<port>.pid record via run_marker. The inline app mint does not. So the fold is a behaviour change to kirocrew app and needs its own review. Carried from span f088218775eb.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=917896ee9da2 rebutted: maintainer ruling on the agent a nameless spawn runs, with an override requested for this head

The item clears on a merged RFC or on a maintainer's /ai-review override first-principles. The maintainer decision is already recorded on this PR: acceptance 5964899894 ("A spawn that names no agent runs kirocrew on every surface"), and overrides 5964017525, 6038349655 and 6048628725 on earlier heads, all for this same item. The decision: the agent governance checks at admission and at the run-time re-check must be the agent that runs. Without the pin, a scopes.agents ceiling is checked on one name while session.pool_agent or agent.default_agent runs another. An override for this head has been requested from the maintainer.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=a7023f87ce8e rebutted: the poll retry, outcome exit and tool-approval hint are the restored blocking command's contract

On main, kirocrew spawn run exited 0 for a stopped run, so a script chaining on it read a false success. Breaking declares the nonzero exit, pinned by test_a_stopped_run_is_not_reported_as_success. The bounded retry (2 attempts) covers only answers the gateway marks retryable: a 5xx with Retry-After, or a GET whose answer was lost. Without it, one such answer ends the blocking wait while the run goes on, which invites a duplicate spawn (test_the_poll_asks_again_after_a_transient_failure, test_the_poll_gives_up_after_its_retries). awaiting_tool_approval is read from the gateway's existing pending-approval state. Without it, a run parked on a tool prompt sits on "waiting for result..." with nothing the user can act on (test_a_run_parked_on_a_tool_prompt_says_so_on_both_read_paths). Ruling 6048628725 covers these. Carried from span 189c919dfa0b.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=bb4d2d1dc03e rebutted: maintainer ruling, a nameless spawn runs kirocrew on every surface

The maintainer decided this on this PR: acceptance 5964899894, plus rulings 5964017525, 5977964351, 5988267324, 6007795939, 6038349655 and 6048628725. The pin is what makes the CLI fix's governance sound. A nameless kirocrew spawn run checked against a scopes.agents ceiling has to be checked as the agent that will run it, or it is checked as one agent and run as another. Operators who relied on agent.default_agent name the agent, and Breaking states that remedy. Carried from span 8d86a6682361.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor
  • span=e9339bdb855b rebutted: governed like the Spawn button, a terminal spawn would escape every surface:cli profile

Rulings 5988267324 and 6007795939 record this decision. The Spawn button's empty key maps to surface unknown (sel._infer_source), and no surface-bound profile binds that surface. Governed the same way, kirocrew spawn run would escape an operator's surface:cli profile, which already binds kirocrew chat and kirocrew computer call. "On main a CLI spawn already ran under the empty key" is true only because on main every CLI spawn got a 409 and never ran.
Evidence: test_a_cli_profile_governs_kirocrew_spawn_run and test_a_retried_cli_run_is_still_governed_as_the_cli go red without the CLI branch of subagent.spawn_governance_key. derive_execution clears ExecutionContext.origin, so no child inherits it. Carried from span 722205b4a3f4.
This head (ddd2cd7) is a maintainer rebase of 37e4d64 onto main aee834c. It applied with no conflicts, so no code changed and the finding is unchanged.

@bolichen97

Copy link
Copy Markdown
Contributor

/ai-review override first-principles ddd2cd7: Maintainer decisions are already recorded (5964017525, 5964899894, 5977964351, 5988267324, 6007795939, 6038349655, 6048628725). A nameless spawn runs kirocrew on every surface, CLI spawns are governed as surface cli, and the app mint fold is deferred to #16350. This head rebases 37e4d64 onto main with no code change. Every FP item repeats one already ruled on.

@github-actions

github-actions Bot commented Oct 8, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@bolichen97 marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for ddd2cd7ca39b0316301dd13664e899f9236a3cb1.

Maintainer decisions are already recorded (5964017525, 5964899894, 5977964351, 5988267324, 6007795939, 6038349655, 6048628725). A nameless spawn runs kirocrew on every surface, CLI spawns are governed as surface cli, and the app mint fold is deferred to #16350. This head rebases 37e4d64 onto main with no code change. Every FP item repeats one already ruled on.

This decision applies only to this commit. A new push requires a new judgment.

@iamwhatever

Copy link
Copy Markdown
Contributor

This makes kirocrew spawn run start a subagent again instead of always failing. It does so with a new owner-gateway client that mints an owner token after checking that the peer is the host's own gateway (run_marker), and by governing CLI spawns as surface cli with a pinned agent. The checks are green at ddd2cd7, and the Design and First Principles items each have a disposition from maintainer bolichen97 citing recorded rulings, plus a first-principles override for this head. I did not try to break the owner-token mint and peer check end to end in this pass. HOLD because the PR touches token minting, dashboard token auth, identity and spawn governance, and it is very large (+2433/-418 across 47 files, with a declared Breaking change for nameless spawns). It needs a maintainer's approval, not an automated one. — crew-pr-reviewer

@koenvincent

Copy link
Copy Markdown
Contributor Author

@bolichen97 Rebased onto current main to resolve the conflict; the only change beyond the base is the MCP route-table fingerprint in test_dashboard_server_composition_contract.py (245 rows, the two /api/token/local routes on top of main's 243). The First Principles findings repeat the items you ruled on for ddd2cd7; could you re-record the override for 03c5118?

@koenvincent

Copy link
Copy Markdown
Contributor Author

@bolichen97 Rebased onto current main (54b0c64) to clear the conflict; head is now e889fde, still one commit.

Beyond the rebase:

  • test/test_spawn_cli_owner.py: the eight asyncio.wait_for(..., timeout=10) waits on run tasks now use await_until, so the wait-discipline ratchet stays at 0 for this new file.
  • dashboard/handlers/messaging.py: one import moved into isort order (the earlier Backend Lint & Type Check failure).
  • test_dashboard_server_composition_contract.py: MCP route-table fingerprint is now 285 rows, main's 283 plus the two /api/token/local routes.
  • test_subagent_continuable.py: kept main's await_until in the two conflicted hunks and kept this PR's admission-time refusal assertions.

The other red shards on the previous head (test_cancellation_during_the_spawn_still_reaps, the ratchet baseline) came from the base and are fixed on main. Locally: test_subagent_continuable.py 202, test_dashboard_server_composition_contract.py 109, test_spawn_cli_owner.py 59, and test_wait_discipline_ratchet.py 39 all pass. isort, flake8 and mypy are clean on the changed files.

The First Principles findings should match the ones you already ruled on. Could you re-record the override for e889fde?

@koenvincent

Copy link
Copy Markdown
Contributor Author

Rebased onto current main (43cc6d3) because it had become CONFLICTING. New head a6e9e4c, still one commit.

The First Principles override was recorded for an older head, so it has to be recorded again for this one. Since #18468 now admits the session-less internal caller, a maintainer may also want to confirm that the owner-token path here is still the route they want for kirocrew spawn run.

@koenvincent

Copy link
Copy Markdown
Contributor Author

Follow-up at 8243257: test_kirocrew_spawn_end_to_end still asserted that the CLI's old shape (internal secret plus X-Internal-Caller, no session and no owner token) is refused with 409. Since #18468 that shape is admitted on main, so the last block of the test contradicted main, and it is removed together with its now-unused _raw_unix_post helper. The owner-token mint, spawn and list, the clock jump and the logout stay in the test. It runs only on a host that verifies the test process as the local owner, so CI is where it executes.

This is worth a maintainer's look before reviewing further: with #18468 merged, kirocrew spawn run reaching a running gateway with only the internal secret already works on main. What this PR still adds is governing those spawns as the attended cli surface through the owner-token path. If that is not wanted, I can close this PR.

kirocrew spawn run posted /api/spawn with the internal secret and no
session. api_spawn claimed the body default "" while the verified
scope had session None, so every CLI spawn got 409
member_identity_unavailable, and a headless gateway could not start a
subagent at all.

The CLI is the dashboard owner at a terminal, so it now authenticates
as the owner principal: it mints a short-lived owner token from
/api/token/local with the local secret, over the dashboard unix socket
only, and sends it on every spawn call. The owner arm of the spawn
routes admits it as it admits the owner's browser; the shared identity
check for internal callers is unchanged.

The spawn carries origin "cli", accepted from the owner principal
only and kept on the task row. Nothing inherits it: a run that run
spawns and a schedule it creates do not carry it. Such a run is
governed as surface cli on every agent name it answers to, the
explicit template and the agent it resolves to, at admission and at
the run-time re-check, and faces the ordinary spawn prompt like any
other caller's. A continuation of it that names a parent session is
governed under that session.
An internal caller with no session now owns no run, parentless ones
included, and its refused spawn is audited as before.

A spawn that names no agent is checked as kirocrew and runs kirocrew
by name, so the agent checked is the one that runs. The run-time
re-check now covers a run that names its agent too, so a scope
tightened while it waited for approval still binds it. The warm pool
matches on the agent that runs, so a pool filled for no agent still
serves a spawn that asks for kirocrew by name. The poll asks
again, a bounded number of times and after Retry-After, when the
gateway answers a 5xx or the answer to a poll is lost; any other
failure ends it with the gateway's or the transport's error. It
exits nonzero for any ending but completed: a stopped run, and a run
whose ending nothing recorded, are not reported as a success. A
gateway older than the CLI sends no outcome, and there an error still
fails and anything else prints the result. Both servers name the unix
socket for the bound port, so --port auto works.

Co-authored-by: Bolin Chen <bolichen@amazon.com>
@koenvincent

Copy link
Copy Markdown
Contributor Author

Rebased onto main c46a1dc (new head 502e9ad, one commit, verified).

  • Conflict: only the MCP route-table fingerprint. It is now 289 rows: main's 287 plus this PR's two /api/token/local rows.
  • First Principles blocker (duplicated peer check): fixed. owner_gateway_client.gateway_peer_verifier and memory_forget_client._gateway_peer_verifier both call one predicate now, run_marker.is_recorded_gateway(peer, record). Each client still raises its own error type and message. A new test covers each refusal case (unknown peer, no record, empty start token, other pid, recycled pid). When I dropped the start-token check from the predicate, both that test and test_a_recycled_gateway_pid_receives_nothing went red.
  • Tested locally: test_memory_forget_cli.py, test_spawn_cli_owner.py and test_dashboard_server_composition_contract.py (202 passed, 1 skipped, plus the new test). black, isort, flake8 and mypy report nothing on the touched files.

The scope question from my earlier comment still stands. Since #18468 merged, kirocrew spawn run works on main without this PR. What remains here is the cli surface governance through the owner-token path. If maintainers don't want that as its own change, I'll close this PR.

@koenvincent

Copy link
Copy Markdown
Contributor Author

Closing this in favour of #18468, which merged on 2026-10-13. With it, kirocrew spawn run starts a subagent again on main, and that was this PR's goal.

What is left here does not stand on its own. First Principles is blocking on the nameless-spawn default agent (a product-shape change without an RFC). The remaining items are a second token-mint path next to app_lifecycle_client and two riders (the tool-approval hints and the poll retries). If maintainers want the cli surface governance (origin=cli on the owner token) as its own change, I'll open a small PR for just that, limited to CLI-origin runs and reusing the existing mint. The shared peer-check predicate from the last push (run_marker.is_recorded_gateway) can come along with it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

fork Pull request from a fork (external contributor) merge conflict Branch has merge conflicts with its base — author must resolve before merge

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants