Skip to content

fix(subagents): grow safe concurrency and support useful delegation - #12203

Merged
bolichen97 merged 1 commit into
mainfrom
fix/adaptive-concurrency-climb
Sep 20, 2026
Merged

bolichen97 merged 1 commit into
mainfrom
fix/adaptive-concurrency-climb

Conversation

@bolichen97

@bolichen97 bolichen97 commented Sep 20, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

A configured limit of 64 can spend a long time at a handful of subagents, or remain near one after pressure. Recovery required twenty completed runs at each step, even when the current cap was one. The model-facing instructions also disagreed with the solo gate: they suggested context isolation as a general excuse, yet could not express useful parent-plus-child work. A 100-tool-call default can cut off a longer assignment before its three-hour execution deadline.

Why it matters

Independent work should begin promptly when the machine can support it. Raising the limit must not spend memory already promised to cold processes, and delegation should reduce useful work time rather than add a worker that merely relays the answer.

What changed (motivation → approach → change)

Case Before After
Parent implementation + one docs child No matching solo reason Explicit reason; parent may do one short separate step
Cold workers awaiting RSS growth Same free memory could be spent repeatedly Pending startup memory is reserved
Long task at cap one Wait for whole task completion Fresh progress can open one trial slot
Work needing more than 100 tool calls Default turn budget ends the run Default 1000 calls, still bounded by three hours

adaptive/policy.py keeps the configured limit as a ceiling. Successful execution earns startup doubling every clear five-second window. After pressure, one-slot recovery needs min(20, current capacity) successful completions. A long task can open one trial slot when it shows fresh progress, work is queued, memory is available and the clear window has no throttling. Stale activity earns nothing; the initialization threshold stays unchanged. Spawn spacing defaults to 0.25 seconds. Disabling adaptive_slow_start disables doubling, not the shorter recovery threshold. config/loader.py, config-baseline.json, owning specs and src/kiro_crew/docs/dynamic-subagent-sizing.md describe these choices. resource_status.py, dashboard/server.py and context.py distinguish effective capacity from the configured ceiling. constants.py gives config, loader, manager and tool descriptions one 1000-call default; the existing three-hour deadline stays. Missing turn-budget keys adopt it automatically. config/superseded_defaults.py reports stored 100 values without guessing whether they were explicit choices.

subagent.py and subagent_manager/admission/gate.py account for resident workers before combining remaining memory slots with total CPU capacity. A configured limit such as 64 is no longer clipped to the automatic-sizing maximum of 32. Admission reserves expected cold-process memory until RSS accounts for it. That reservation cannot be negative, even with a misconfigured cost. Windows/macOS use native readings; Linux respects process cgroups and visible ancestors. A finite container limit with unreadable usage establishes no spare capacity. Host-read failure retains a known container bound; absent bounds retain the existing fallback. Estimates cannot guarantee safety against arbitrary later allocations.

config/prompt.md, the Autopilot prompt and mcp_tools/spawn.py favor focused direct work and bounded, ready delegation with clear ownership and verifiable output. solo_spawn.py and validation.py add parent_parallel, specialist and user_requested, requiring concrete solo_details. subagent_persistence.py retains redacted model claims, not proof or authorization. Async spawn tells the parent when it may do one separate step of at most a minute, then yield. This applies only to dashboard-owned slots that can deliver child completion while the parent is busy. Blocking spawn rejects parent parallelism. Existing reasons and programmatic payloads, batch barriers, cancellation, retry/continuation and approval boundaries remain. Tests cover these contracts, early completion, startup memory and long-task recovery. eval/scenarios/subagent_policy.json supplies eighteen decision cases without claiming executed children or measured speedup.

Decision context: overload RFC, amended here to change the previously accepted defaults, and requester-authorized scope. The explicit policy decision accepts these RFC changes and run-local audit claims. The requester authorized the finite long-task default and conservative upgrade policy.

Backwards compatibility

Breaking: with the memory floor enabled, admission may now defer/refuse a start whose pending process memory would consume that floor; native Windows/macOS starts honor it too. A finite Linux container limit with unreadable usage defers starts while spare capacity is unknown. Setting spawn_min_memory_gb=0 disables both this floor and the cold-start reservation; the faster ramp can then commit memory before reactive pressure cuts catch it. The default keeps this guard enabled. Writer sweep on origin/main @ 2f11434bdccebd19e7d5020ca2bb19f731436f13: MCP spawn tools; dashboard spawn/retry/resume; messaging commands and gateway callers; spawn_warm; manager admission, continuation and cancellation; apps spawn SDK, Code Review Sage and Mochi; and workflow-memory tests. They retain the existing queue/refusal handling. No existing reason requires a new field; solo_details is required only for newly added reasons. Reason text never grants permission. The turn-budget default changes only for unset configurations: stored values, including 100, remain authoritative, and overlays/per-spawn overrides keep precedence. Older full-config saves erased default-versus-user provenance, so those ambiguous 100 values need explicit adoption through the existing defaults command. Historical timeout migrations are unchanged.

Tests

Local validation and failure/recovery provenance account for the resolved 52-gate floor and all 2,194 related backend files. Complete affected-file reruns resolve the recorded fixture failures; affected-file recovery verification passed 1,212 Linux tests. Resource-floor regressions and related admission/controller tests pass 655 Linux tests. Base integration verification passed 549 backend and 104 frontend tests, plus Python and TypeScript checks. Frontend related tests passed 24,649 with two skipped; Electron passed 1,959 with three skipped. Static, type, documentation, build and bundle checks passed. The format gate prunes two clean files from .github/black-baseline.txt.

test/test_spawn_solo_gate.py and related tests cover solo reasons and bounded parent work, metadata round trips, queue/early completion, delayed startup memory, native/cgroup admission and progress without completion. The finite-cgroup/unreadable-usage regression failed 32 cases before repair. Negative-cost regressions failed three cases before repair and preserve the floor even on an exhausted cgroup; the complete startup-memory file passes 33 Windows tests. Windows affected-file runs passed 245 tests; sizing/coverage passed 366 with five expected failures. The 101-tool manager regression fails on the old default and completes on the new budget; explicit stored caps and overlays survive load/save/reload. The defaults-focused Windows run passed 86 with three platform skips. Linux verification passed 1870 tests across the sixteen related files on this head; all forty Python/static/documentation gates passed, including mypy across 1654 source files. The frontend and Electron delta from the prior green head is empty. Full platform suites and actual remote review remain CI's authority.

Manual verification

Real manager/pump/controller fault injection uses virtual time and delayed synthetic RSS, not large host allocations. With a post-sample memory shock, disabling the reservation exhausts memory after 18 starts; the repair stops after 10 and preserves 4 GB. An ample-host case still starts 18 workers quickly and leaves 16 GB. These results bound the tested cost estimate, not every possible workload.

Decision-only before/after traces and their limits were collected with tools disabled. They are not end-to-end child execution or performance measurements; served model and token counts are unverified/unknown. Repeat evaluation remains incomplete after the local CLI became unavailable. Local review used GPT 5.6 and a GPT-family fallback applying the Opus contract; the remote Opus lane remains required.

Related Issues

no linked issue: this PR completes the requester-authorized subagent follow-up and preserves the merged earned-success fix from #12167.

Pattern harvest

Rule candidate: review-prompt
Pattern: distinguish remaining headroom from total capacity, and account for admitted work before its resource use becomes observable.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated
  • No secrets, credentials, or internal references in the diff

@bolichen97
bolichen97 requested review from a team and iamwhatever September 20, 2026 05:28
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 20, 2026
@github-actions

github-actions Bot commented Sep 20, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of b6e60cb6bbdce9453562bf16dcc29c97e80a2e67 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] b6e60cb

False positive or not applicable? A repository writer can comment:
/ai-review override gpt b6e60cb6bbdce9453562bf16dcc29c97e80a2e67: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 20, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ human override accepted

@bolichen97 overrode this lane for b6e60cb6bbdce9453562bf16dcc29c97e80a2e67 via /ai-review override (this commit only).

@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 20, 2026
@bolichen97
bolichen97 force-pushed the fix/adaptive-concurrency-climb branch from c72bebb to 276b877 Compare September 20, 2026 05:42
@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 20, 2026
@github-actions

github-actions Bot commented Sep 20, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of b6e60cb6bbdce9453562bf16dcc29c97e80a2e67 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Verify the now-enforced spawn admission floor on a real macOS host; the PR's cited verification runs cover Windows and Linux only.

[DESIGN-REVIEWED] b6e60cb

@github-actions

github-actions Bot commented Sep 20, 2026 •

Copy link
Copy Markdown
Contributor

Opus 4.8 Review — ✅ no blocking findings

Reviewed b6e60cb6bbdce9453562bf16dcc29c97e80a2e67 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] b6e60cb

Verdict parsed from the review's SHA-scoped output markers for commit b6e60cb6bbdce9453562bf16dcc29c97e80a2e67.

False positive or not applicable? A repository writer can comment:
/ai-review override fable b6e60cb6bbdce9453562bf16dcc29c97e80a2e67: <one-sentence reason>

@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 20, 2026
@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: no
mechanism: slow start, a default-on x2-per-5s growth regime replacing +1-per-30s

  • needs-a-decision: the x2 default regime is not the accepted RFC's recorded rule span=fa35e879e3b9

The item is correct and I cannot clear it myself. rfc-overload-resilience
records "+1 after >=20 successful inits AND >=30s"; this PR ships x2 per 5s
clean window as the DEFAULT until first corroborated pressure. Whether the
RFC's "Thresholds above are starting values" license covers a change of
regime, rather than a change of numbers, is a maintainer call.
The defect it addresses is real and measured: a decrease halves on one
corroborated sample while an increase needed ~60 clean windows plus ~1200
finished runs to cross a 64 ceiling, and sub-agents themselves produce the
loop lag that cuts, so a cap that fell to the floor did not come back.
Maintainer: rule on ONE of (a) ship x2 as default, (b) ship it opt-in with
adaptive_slow_start defaulting false, (c) amend the RFC first.
I will not self-authorize the override this item names.

@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: no
mechanism: spawn-stagger default lowered 2.0s -> 0.25s for every install

  • needs-a-decision: the 8x stagger cut is a default every install gets span=facd0c5a7485

The item is correct on both halves. The accepted RFC's limiter table records
2.0s, and the harm this PR names -- "a 64-slot cap took over two minutes to
fill" -- is reachable only at a high explicit max_subagents; the shipped
default is 0 (auto), which on most hosts resolves far below 64, so the
named harm is not what a default install experiences.
The change was made at the repository owner's explicit request in this PR,
which is why it is here rather than rebutted or dropped.
One more fact for the ruling: lowering the interval makes provider-side
throttling MORE likely (AcpError waves have been observed past ~5 concurrent
dispatches); the controller does observe per_provider_429 and backs off.
Maintainer: rule on ONE of (a) keep 0.25 as the default, (b) a middle value
such as 1.0, (c) leave the default at 2.0 and carry the cut in its own PR.

@bolichen97
bolichen97 force-pushed the fix/adaptive-concurrency-climb branch from 276b877 to 7069db2 Compare September 20, 2026 06:36
@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: passed Eligible automated validation passed for the current revision readiness: checking Automated validation is still running labels Sep 20, 2026
@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: no
mechanism: adaptive growth regime plus spawn-fill default, both differing from the accepted RFC's recorded shape

  • needs-a-decision: the new growth and stagger defaults are not the shape the accepted RFC records span=afce1f14966a

The item is correct and no writer on this PR can clear it. rfc-overload-resilience
is accepted on the base branch and its §5.2 table records +1 after >=20 successes
plus a 2.0s spawn stagger. This head ships x2 per clear 5s window until first
pressure, an exec bar of min(increase_successes, cap), and a 0.25s stagger.
Those are default-behaviour changes for every install, not parameter retunes, so the
RFC's "the thresholds above are starting values" sentence does not cover them.
The maintainer has two paths and only he may pick: merge an RFC amendment on main
recording the new growth and stagger defaults, then rebase this branch; or record
/ai-review override first-principles 7069db257faa562ef5cc3c25480aee65a9bd7e6c: <decision>.
Not self-authorized here.

@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: no
mechanism: slow start, a default-on x2-per-5s growth regime replacing +1-per-30s

  • needs-a-decision: the x2 default regime replaces the RFC's recorded +1 rule span=b8860b8d51c4

Correct as stated. The regime change, not its numbers, is what the accepted
RFC does not record, so this cannot be rebutted as a retune. It also cannot be
deferred: the default ships with the diff.
Waiting on the maintainer decision named in the Blockers disposition
(span=afce1f14966a). The option I would take is keep slow start default-on,
since the defect it fixes is measured (cap pinned at the floor and unable to
climb back), and record that decision in the RFC rather than in review.

@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: no
mechanism: exec increase bar changed from a flat 20 successes to min(increase_successes, cap)

  • needs-a-decision: the min(increase_successes, cap) bar has no opt-out span=4d43b97696ff

Correct: there is no flag that restores the flat 20, so every install gets the
new bar. The reason it has no opt-out is that a flat 20 at cap=1 requires twenty
serial runs to earn cap=2, which is the pinned-at-floor defect this PR exists to
remove; an opt-out would preserve the defect behind a switch.
That reasoning is a product ruling, not a review one, so it is the maintainer's.
Waiting on the decision named in the Blockers disposition (span=afce1f14966a).

@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: no
mechanism: spawn-stagger default lowered 2.0s -> 0.25s for every install

  • needs-a-decision: the 8x stagger cut is a default every install gets span=ae07aa0674ae

Correct on both halves, and the second half is the sharper one. The accepted
RFC's limiter table records 2.0s, and the harm this PR names -- "64 slots take
over two minutes to fill" -- only holds when max_subagents is pinned high. The
default is 0 (auto), which sizes most hosts well below 64, so a default install
does not experience the harm the change is justified by.
The cut is in this PR because the maintainer asked for it, which is exactly why
the ruling is his. Waiting on the Blockers disposition (span=afce1f14966a).
The option I would take is 1.0s: it removes the two-minute fill without an 8x
move away from a recorded default.

@bolichen97
bolichen97 force-pushed the fix/adaptive-concurrency-climb branch from 7069db2 to 8c13b97 Compare September 20, 2026 07:31
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 7069db257faa562ef5cc3c25480aee65a9bd7e6c; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 7069db257faa562ef5cc3c25480aee65a9bd7e6c: <one-sentence reason>

@bolichen97

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles 8c13b97: Maintainer decision (bolichen97), recorded on the agent's behalf with explicit authorization. All three default changes ship as-is: slow start default-on (x2 per clear 5s window until first corroborated pressure), the exec bar at min(increase_successes, cap) with no opt-out, and subagent_spawn_stagger_secs at 0.25s. Rationale: the pair the RFC recorded cannot reach a user's configured ceiling — a x0.5 decrease fires on one lag spike while a flat 20-success bar at cap 1 needs twenty serial runs to earn cap 2, and a 2.0s stagger takes over two minutes to fill 64 slots even once the cap allows them. An opt-out for the old bar would preserve the defect behind a switch. The climb is bounded by the host's memory and CPU terms, so no default can exceed the machine. rfc-overload-resilience.md is amended in this PR to record the shipped defaults in §5.2 and the limiter table, so the decision lives in the RFC rather than in review; the lane reads the base ref, which carries the amendment only after this PR merges.

@github-actions

Copy link
Copy Markdown
Contributor

AI-review override not recorded: keep the reason to 500 characters or fewer.

@bolichen97

Copy link
Copy Markdown
Contributor Author

self-added: yes
mechanism: _slow_start_retired, a per-process flag separating a config-off slow start from one retired by observed pressure

  • fixed: re-enabling the flag before any pressure now restores slow start span=2ceb54b791dc

Confirmed in code. update_params set _slow_start = False on a config-off and
never restored it, so off-then-on left slow start dead for the process even when
nothing had been observed. The docstring justified one-way as "that evidence does
not expire with a config edit" -- which does not cover a process where no evidence
exists; the code conflated "config said off" with "pressure retired it".
Fixed in 8c13b97: a _slow_start_retired flag is
set at both retirement sites (corroborated pressure, pause), and update_params
computes slow_start and not retired. Three regression tests cover off->on with no
pressure (restored), pressure-then-off->on and pause-then-off->on (stay retired).

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Sep 20, 2026
@bolichen97
bolichen97 enabled auto-merge (squash) September 20, 2026 16:47
@bolichen97
bolichen97 merged commit 4a94595 into main Sep 20, 2026
95 of 96 checks passed
@bolichen97
bolichen97 deleted the fix/adaptive-concurrency-climb branch September 20, 2026 16:52
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Sep 20, 2026
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.
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