Skip to content

feat(spawn): refuse a one-task spawn that gives no solo_reason - #11710

Merged
chenmingwei23 merged 1 commit into
mainfrom
feat/spawn-single-task-gate
Sep 19, 2026
Merged

chenmingwei23 merged 1 commit into
mainfrom
feat/spawn-single-task-gate

Conversation

@iamwhatever

@iamwhatever iamwhatever commented Sep 18, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Decision recorded in docs/request-for-change/rfc-solo-spawn-gate.md (merged in #11848, status in-progress, implementation PR: this one).

spawn_run(task=...) with ONE task spawns one sub-agent for one job. That is a round-trip with no parallelism gain. The tool text already says "do it yourself", but text is advice: the base prompt's "delegate to a sub-agent to preserve context" kept winning, and a lone sub-agent kept appearing for a single investigation, fix or review.

Why it matters

Every lone sub-agent is a full extra model run for work the parent could do in its own turn. The rule had to be re-taught by hand, and a rule that lives only in prose does not hold.

What changed (motivation → approach → change)

A one-task spawn is now a handshake, not a straight dispatch. The tool refuses the first call and asks: can you do this yourself? The second call must carry a reason from a closed list, or name a model / agent / crew that is really not the caller's own. The refusal names the solo_reason parameter and points at its description; it never spells the passing words, which live in the schema alone.

Call Before After
tasks=[a, b] (2+ tasks) 🟦 spawns 🟦 spawns
one task, nothing else 🟥 spawns 🟩 refused with the question; nothing spawned
one task + solo_reason (bulk_data / fresh_context) — 🟩 spawns; reason in result line and audit
one task + a model / agent / crew that differs from the parent 🟦 spawns 🟨 spawns; audit records what differed
one task + the parent's OWN agent / model / crew 🟥 spawns 🟩 gateway refuses: solo_spawn_unjustified
one task + only keep=true 🟥 spawns 🟩 refused (keep is not a reason)
SDK / app POST to /api/spawn 🟦 spawns 🟦 spawns (never gated)

🟩 added · 🟨 changed · 🟥 removed · 🟦 unchanged

Why a closed list and not a yes/no flag: a confirm=true is rubber-stamped by the second call. The list has no "preserve my context" and no "separate investigation", so the caller has to pick a reason it can defend. bulk_data means the step would flood the caller's context with output. fresh_context means the result would be wrong if the run saw this session's context (a blind review, a clean-slate repro).

Why two halves: the MCP tool process is the only place the task COUNT is known (the gateway sees one POST per task), so mcp_tools/spawn.py refuses the one-task call that names nothing. Only the gateway knows the parent session's own agent and, for a dashboard slot, its model, so dashboard/handlers/messaging.py catches the call that named the parent's own agent / model / crew to slip past. agent is compared against the parent's resolved template, never its member alias, so a member on kirocrew-worker cannot name kirocrew-worker as "different". The tool marks such a POST with solo=true; the SDK and apps never send it, so they are never gated. An unknown parent fact fails open: the roster check never refuses what the tool side let through unless it can prove a match. All three gate outcomes are audited as spawn.solo: denied, allowed on a reason, allowed on a difference (with what differed, e.g. differs=agent).

solo_spawn.py (new) holds the vocabulary, the refusal text, the result line and the roster comparison, so the tool description, the validator and the gate cannot drift apart. validation.py bounds solo_reason to that vocabulary on both SPAWN_RUN_SCHEMA and SPAWN_SUB_AGENTS_SCHEMA. spawn_sub_agents gets the same gate: one entry, no agent_or_mode, no reason → refused.

Words the agent reads: config/prompt.md step 1 of Subagent Orchestration names the refusal and the two reasons; docs/subagents.md gains a "The solo gate" section; the module spec docs/system-specs/modules/subagent.md describes the gate where it pinned the old one-task shape; the Dev Fleet pod-e2e/SKILL.md passes solo_reason="bulk_data" for its one QA sub-agent and stops calling "not derailing my context" a reason to delegate. Every shipped instruction that produced a bare one-task spawn is brought under the gate: builtin_skills/ops-mission-control/sops/dispatch.md (the unattended investigator launches with fresh_context), builtin_skills/frontend-design-workflow/SKILL.md (the blind usability reviewer is a fresh_context case and now says so), builtin_skills/image-authoring/SKILL.md (fan out only for 2+ images; one image is authored inline). The offline E2E stub scenario testing/workflow_memory_scenario.py passes the same reason for its nested one-task spawn, which exercises private-memory routing, not the gate.

Tests

test/test_spawn_solo_gate.py, 53 tests:

  • tool side: one task with nothing → refused before any POST, audited as a refused tool call and as spawn.solo denied; each reason → spawns with the reason on the wire and in the result line; a named model / agent / crew → passes and the result points at the spawn.solo audit for the ground; model="auto" (the "no model chosen" sentinel) counts as nothing named and is refused; keep alone → refused; 2+ tasks → never gated; the same for spawn_sub_agents with one entry.
  • roster check: own agent, own model, a template session naming the default crew, a member naming itself as crew, a member naming its own resolved template as agent → ""; another agent / model / crew → the ground (agent / model / crew); a dashboard slot is found by its effective session key, so a chat linked to a channel thread is compared too; unknown parent, unpinned slot model, a slot owned by another session → fail open and say (parent unknown).
  • gateway: solo=true naming the parent's own agent → 400 solo_spawn_unjustified, audit spawn.solo denied; a member naming its own template → the same 400; a reason → allowed and audited; another agent → allowed and audited with differs=agent; a POST without the solo marker → never gated; a reason outside the vocabulary → validation 400.
  • schema: the vocabulary is closed, "" means not given, and both tools advertise the enforced gate with the same enum.

Nine existing spawn test files that drove OTHER behaviour through a lone task (effort forwarding, the context-group wire contract, cancellation, audit owner, redaction, model forwarding) now pass solo_reason="bulk_data" so they still reach the POST they assert on; test_spawn_reasoning_effort.py and test_subagent_context_group_boundary.py do it through a small _through_solo_gate helper that adds the reason only where the gate would fire.

Manual verification

N/A — unit coverage sufficient: both halves of the gate are pure functions over the call's arguments and the parent's roster, and every row of the matrix above is a test.

Related Issues

no linked issue: the gate came out of a dashboard conversation, not a tracked issue.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|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)
  • No secrets, credentials, or internal references in the diff

@iamwhatever
iamwhatever requested review from a team and bolichen97 September 18, 2026 06:17
@iamwhatever

Copy link
Copy Markdown
Contributor Author

Intent: Make the "one sub-agent for one task is not a reason" rule mechanical: a one-task spawn_run / spawn_sub_agents call that names no solo_reason and no model/agent/crew other than the caller's own is refused with a question instead of spawned, so a lone sub-agent only ever appears on a defensible, audited ground.
Not a goal: Gating batches (2+ tasks), the SDK or apps posting to /api/spawn directly, or verifying the truth of a caller's solo_reason -- the gate forces the decision to be made and recorded, it does not adjudicate it.

@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 18, 2026
@github-actions

github-actions Bot commented Sep 18, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — 🟡 CONCERNS

Design-level review of c9af747878133d247d301c0130ed257ae097881d — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

Closed vocabulary omits the user-requested background job, so honest models block the chat slot or file false bulk_data claims that corrupt the audit signal.

Watch

  • RFC open question 1 ships unresolved with a lossy default: "run this one long job while we keep talking" fits neither reason, so the model either runs it inline (blocking the slot the user is talking in) or retries with a false bulk_data — which is exactly the "uniform bulk_data retries" pattern the RFC's success criterion treats as failure, polluted from day one by a legitimate case.
    Clears when: open question 1 is decided (a long_runningreason, or the spec states this case runs inline by design), or post-mergespawn.solo data shows denials converting to inline work.

Suggestions

  • config/prompt.md step 1 spells the two passing tokens bare (bulk_data / fresh_context), undoing the refusal's deliberate "never spell the passing words" design — name only the solo_reason parameter there so a retry must read the glosses.

[DESIGN-REVIEWED] c9af747

@github-actions

github-actions Bot commented Sep 18, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of c9af747878133d247d301c0130ed257ae097881d and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] c9af747

False positive or not applicable? A repository writer can comment:
/ai-review override gpt c9af747878133d247d301c0130ed257ae097881d: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 18, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of c9af747878133d247d301c0130ed257ae097881d — 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: PASS

Verify model-pinned skill spawns (prepare-pr retrospective, llm-council members) survive a parent slot pinned to the same model — the gate refuses that match by design.

What this change ships

Inventory (10 items) — 10 justified

Intent: stop the agent spawning one sub-agent for one task it could do itself, by making the tool refuse until a defensible reason is given — an ADDITION, implementing the decision recorded in rfc-solo-spawn-gate.md (in-progress on the base list, text covers this exact shape including the closed vocabulary, both halves, fail-open, and the skill updates).

  1. A bare one-task spawn_run / one-entry spawn_sub_agents is refused with a question; nothing spawns — justified
  2. New solo_reason parameter, closed enum bulk_data/fresh_context, on both spawn tools — justified
  3. Gateway returns 400 solo_spawn_unjustified when a solo call names the parent's own agent/model/crew — justified
  4. Every gate outcome is audited as spawn.solo (denied / reason / difference) — justified
  5. Spawn result line now shows the reason or points at the audit — justified
  6. Base prompt step 1 steers to tasks=[…] and names the refusal — justified
  7. Four shipped skills/SOPs now pass a reason or fan out only for 2+ — justified
  8. keep=true alone no longer opens the gate — justified
  9. Spec and user docs gain "The solo gate" in the same commit — justified
  10. Offline E2E stub's nested one-task spawn passes bulk_data so it keeps exercising memory routing — justified

Checks run: spawn_run(task= in src/ — 3 hits, all gate-aware after this diff; remaining one-task instructions (pipeline-conductor inspector, prepare-pr cold reader/retrospective, llm-council fan-out) all name a different agent or model, so none is a bare unfixed sibling. The old pin test_spawn_single_task_gate.py ("not read as forbidden") pins description wording, is untouched, and the reversal of the advisory stance is the RFC's recorded decision. parent_slot_model has one consumer, in-module.

[FIRST-PRINCIPLES-REVIEWED] c9af747

@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 Sep 18, 2026
@github-actions

github-actions Bot commented Sep 18, 2026 •

Copy link
Copy Markdown
Contributor

Opus 4.8 Review — ✅ no blocking findings

Reviewed c9af747878133d247d301c0130ed257ae097881d — this comment is updated in place on each push.

Review details

The discovery pass produced no candidates, and having traced both halves of the solo gate through the diff — the tool-side count refusal in mcp_tools/spawn.py, the gateway roster check in solo_spawn.solo_spawn_difference / api_spawn, the closed-vocabulary schema bound, and the fail-open branches — I found nothing that grounds a concrete input → call path → wrong-outcome chain at the required bar. The one normalization asymmetry noted (a slot model like "Auto" bypassing _UNKNOWN_MODEL's exact-match set) resolves only in the permissive fail-open direction the module intends, so it is not a defect.

No findings.

[OPUS-REVIEWED] c9af747

Verdict parsed from the review's SHA-scoped output markers for commit c9af747878133d247d301c0130ed257ae097881d.

False positive or not applicable? A repository writer can comment:
/ai-review override fable c9af747878133d247d301c0130ed257ae097881d: <one-sentence reason>

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Gate premise (refusal converts to inline work, not a reflexive reason) is self-certified span=9ba7dad2b94e

disposition: rebutted
Not a defect in the shipped code: the reason is the caller's own claim BY DESIGN (module docstring, "the gate forces the decision to be made and recorded, it does not adjudicate it" in the intent comment).
No pre-merge check can verify a model's future behaviour; the closed vocabulary deliberately omits "preserve context" / "investigation" so a rubber stamp has to be a false claim, not a vague one.
The spawn.solo audit (denied / allowed on reason / allowed on difference) is exactly the instrument the Clears-when names; without the gate the same lone spawns happen with zero visibility.

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Long-running job with small output fits neither reason span=b4866b772724

disposition: needs-a-decision
Maintainer question: should the closed vocabulary gain a third reason (e.g. long_running: the step runs for minutes on its own and the chat slot must stay usable; only the summary is needed), or is running such a job inline the intended default?
Adding it widens the vocabulary the PR was designed to keep small; omitting it makes an honest model block the slot for the job's duration. Decision requested in the PR thread or via a maintainer comment.

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Product-shape change without accepted RFC (not justified as shipped) span=ca01f8297cc3

disposition: needs-a-decision
The default of a first-class tool changes (one-task spawn: spawns -> refused unless justified). No RFC on the base records it; the base spec docs/system-specs/modules/subagent.md is updated in this PR to describe the gate, which is a proposal, not a record.
Maintainer decision requested: record it with /ai-review override first-principles <head sha>: <decision>, or ask for an RFC under docs/request-for-change/ to be merged first (then push or rebase).

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Blocker: one-task-spawn default flips without a recorded decision (items 1 & 6) span=babbb8e62fac

disposition: needs-a-decision
Same decision as the not-justified item: the change is intentional (the gate makes the existing "do a single task yourself" rule mechanical) and item 6 follows from item 1 (the pod-e2e skill's delegate-by-default becomes inline-by-default).
Maintainer decision requested: /ai-review override first-principles <head sha>: <decision>, or an RFC merged to main first.

@iamwhatever
iamwhatever force-pushed the feat/spawn-single-task-gate branch from 7854070 to 50384fb Compare September 18, 2026 06:34
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 18, 2026
@iamwhatever
iamwhatever force-pushed the feat/spawn-single-task-gate branch from 50384fb to 7e169cb Compare September 18, 2026 06:53
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Sep 18, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • if model or agent or crew accepts model="auto", letting a lone task through unjustified span=b71ebb1f8e4b

disposition: fixed in 7e169cb
_named_model() maps the auto sentinel (any case, padded) and empty to "nothing named" at the tool-side refusal, the result line and the gateway roster check, so model="auto" alone is refused before any POST and can never be the ground a spawn is let through on.
Tests: test_the_auto_sentinel_is_not_a_named_model, test_model_auto_alone_is_refused_before_any_post, test_the_auto_sentinel_is_nothing_named.

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • function-local imports violate top-level-imports span=b71ebb1f8e4b

disposition: fixed in 7e169cb
canonical_key is imported at module scope (model_registry depends only on stdlib + atomic_write, no cycle) and the defensive try/except around it is gone.
effective_session_key stays function-local with a # circular import comment, the rule's documented exception: chat_utils imports dashboard.state, which reaches validation.py, which imports this module.

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Product-shape change without accepted RFC (not justified as shipped) span=ab52c008fb36

disposition: needs-a-decision
The default of a first-class tool changes on purpose: a one-task spawn is refused unless justified, which makes the existing "do a single task yourself" rule mechanical. No RFC on the base records it; the base spec docs/system-specs/modules/subagent.md is updated in this PR, which is a proposal, not a record.
Maintainer decision requested on the current head: /ai-review override first-principles 7e169cb9dfd6cc62d812dd29875423b35fb96df7: <decision>, or an RFC under docs/request-for-change/ merged first (then push or rebase).

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Blocker: the one-task-spawn default is replaced with a refusal without a recorded decision span=b417c2ba40d2

disposition: needs-a-decision
Same decision as the not-justified item: intended, on by default, no opt-out beyond a reason or a different model/agent/crew; programmatic /api/spawn callers are unaffected.
Maintainer decision requested: /ai-review override first-principles 7e169cb9dfd6cc62d812dd29875423b35fb96df7: <decision>, or an RFC merged to main first.

@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 Sep 18, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Value hypothesis unverified; the refusal teaches the two reason words span=b4883cfb0778

disposition: rebutted
Same class as span 9ba7dad2b94e on 7854070: the premise (a refusal converts to inline work) is a model-behaviour claim no pre-merge test can verify by construction; the shipped instrument for the Clears-when is the spawn.solo audit this PR adds (denied / allowed-on-reason / allowed-on-difference).
The refusal reveals nothing the caller lacked: both reasons are already advertised in the tool schema's enum and gloss, so a refusal that hid them would only strand the legitimate bulk_data / fresh_context caller. What the vocabulary withholds on purpose is "preserve context" and "investigation", so a retry has to be a false claim, not a vague one.

@iamwhatever

Copy link
Copy Markdown
Contributor Author
  • Roster half fails open off-dashboard; plumb the channel session's model later if audit shows own-model evasion span=4993c963b0b1

disposition: rebutted
Agreed on the facts, and the finding itself conditions the work on audit evidence that does not exist yet ("if audit later shows"); adding model plumbing for channel sessions now is speculative hardening outside this PR's intent (make the existing rule mechanical; fail open on any unknown parent fact, as the module and docs state).
Fail-open is the documented contract, not a gap: an un-comparable value never refuses a spawn the tool side let through. Channel-LINKED dashboard slots are a different case and are fixed on the next head (GPT span b71ebb1f8e4b).

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Sep 18, 2026
@chenmingwei23
chenmingwei23 merged commit 7373317 into main Sep 19, 2026
86 checks passed
@chenmingwei23
chenmingwei23 deleted the feat/spawn-single-task-gate branch September 19, 2026 00:54
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Sep 19, 2026
amadsalmon pushed a commit to amadsalmon/KiroCrew that referenced this pull request Sep 22, 2026
A one-task spawn_run / one-entry spawn_sub_agents that names no
solo_reason and no model/agent/crew other than the caller's own is
refused with a question instead of spawned; the reason is a closed
vocabulary (bulk_data, fresh_context); a genuinely different
model/agent/crew is a reason in itself, checked by the gateway against
the parent's own; every outcome is audited as spawn.solo; unknown parent
facts fail open; programmatic /api/spawn callers are not gated.

This is the product-shape record the First Principles lane requires for
PR kirodotdev#11710, which implements the design whole. Status in-progress; the
README index row states what is on main today (nothing of the design).

Co-authored-by: EC2 Default User <ec2-user@ip-172-31-10-71.ec2.internal>
amadsalmon pushed a commit to amadsalmon/KiroCrew that referenced this pull request Sep 22, 2026
…otdev#11710)

A single spawn_run(task=...) or spawn_sub_agents with one entry is a
round-trip with no parallelism gain, and the textual "do it yourself"
gate in the tool descriptions kept losing to the base prompt's
"delegate to a sub-agent". Make the gate mechanical: a one-task call
that names no solo_reason and no model/agent/crew other than the
caller's own is refused before any POST, and the result asks whether
the caller can do the task itself. A retry needs a reason from a closed
vocabulary (bulk_data, fresh_context) or a genuinely different
model/agent/crew; the gateway compares the named one against the
parent session's own and refuses a match with solo_spawn_unjustified.

The reason is echoed in the spawn result and audited as spawn.solo.
Batches of 2+ tasks and programmatic /api/spawn clients (SDK, apps) do
not pass through the gate. Docs, prompt.md and the pod-e2e skill name
the new parameter; existing spawn tests that exercised other behaviour
through a lone task now pass a reason.

Co-authored-by: EC2 Default User <ec2-user@ip-172-31-10-71.ec2.internal>
iamwhatever pushed a commit that referenced this pull request Sep 24, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 24, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 24, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 24, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 25, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 26, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 26, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 26, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 26, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 26, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 26, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 27, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
iamwhatever pushed a commit that referenced this pull request Sep 27, 2026
…efault in the prompts

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.
chenmingwei23 pushed a commit that referenced this pull request Sep 28, 2026
…efault in the prompts (#13458)

Users report frequent single-child spawn_run calls. The solo-spawn gate (#11710, widened in #12203) validates reason shape, not value, so the model always has a passing token. A local spawn.solo audit (2026-09-22 to 2026-09-24) showed 33 of 34 one-task calls allowed.

- Gate removed: solo_spawn.py, the MCP tool half and the api_spawn roster check/audit are gone. The parent_work_supported receipt check moves to parent_work.py, unchanged.
- Compat: solo_reason / solo_details are no longer advertised. Validation still accepts and ignores them so old skills are not refused as unknown fields.
- prompt.md / prompt-orchestrator.md: a concrete one-task default; the numbered dispatch -> END YOUR TURN -> wait-for-batch steps are back.
- Member protocol (context.py): the front-desk rule does focused work inline; escalation drafts are self-reviewed.
- subagent.md: delegation policy is prompt guidance only.
- Tests rewritten for the new default.

The RFC supersede decision is a separate doc-only PR (#13511) so the First Principles lane reads it off the base branch first.

Co-authored-by: Joe Guo <zejiangg@amazon.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants