Repository navigation
fix(skills): respect disabled approval setting - #7617
NicholasRBowers merged 1 commit into
Conversation
GPT 5.6 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
Opus 4.8 Review (fork) — ✅ no blocking findingsReviewed Review detailsThe working tree is at main (54fa0ed), not the PR HEAD, so the diff file is authoritative for the change. Let me verify the staging branch still emits an observable SEL audit and confirm the case table. The relevant code and diff are clear. Let me trace the candidate's claim carefully. The new logic:
So the mixed candidate stages into pending even with approval disabled. Now testing outcome (c):
The candidate's own confidence is "low" and concedes "this is arguably the intended trade-off." Under falsification, outcome (c) does not hold — the behavior is the documented, audited, intended design. Killed. No further defect exists in the reshaped branch: the case table is complete and consistent (approval on → always stage; prose-only + approval off → auto-publish; all-invalid scripts + approval off → audited reject; mixed/valid scripts + approval off → stage). No crash, no data loss, no removed guard without replacement. No findings. [OPUS-REVIEWED] 047d51c |
First Principles Review (Fable 5, fork) — ✅ PASSPremise-level review of All the facts I need are verified. The base First-Principles-Verdict: PASS Verify one thing: the rewritten "(GPT MEDIUM)" pin — with approval off, the prose learning is now silently discarded (SEL/log only), not queued for salvage. What this change shipsInventory (5 items) — 5 justifiedIntent: stop approval-off users receiving review requests for candidates whose scripts all failed validation — a FIX.
The pin rewrite carries its evidence: the deleted comment justified staging solely as the means to prevent auto-publish, and the new test keeps [FIRST-PRINCIPLES-REVIEWED] 047d51c |
Design Review (Fable 5, fork) — ✅ PASSDesign-level review of Verified the base code matches the diff context; the reject branch consumes exactly the case the removed Design-Verdict: PASS A real UX/trust bug fixed at the decision point, with the original no-auto-publish safety pin preserved and spec, test, and SEL audit updated coherently. [DESIGN-REVIEWED] 047d51c |
be48b6d to
207ccd3
Compare
Open PR relationship auditThis is a consolidated, point-in-time code-level audit note. It compares complete merge-base diffs and current/merged code; it does not treat a shared topic as duplication or partial coverage as completion. Relationship findings
No PR, Issue, label, branch, or review state was changed by the relationship-note portion of this audit. |
|
🤖 Kiro Crew [operator: NicholasRBowers#a942f9ca]: This PR has been inactive for 7+ days with failing CI. I've assessed the blockers and they appear resolvable — I'll push fixes directly to this branch as a co-author. Assessment: Two blockers, both from a stale base: (1) merge conflict with main; (2) Dependency Audit red on high vulns (fast-uri, js-yaml) in If you'd prefer I don't touch this PR, add the |
207ccd3 to
7d356b0
Compare
|
Intent: When skill approval is disabled, an auto-skill candidate whose supplied scripts all fail validation is rejected outright — never auto-published as disguised prose and never staged into a review queue the user opted out of. |
When skills.approval_required is disabled and an auto-skill candidate supplies scripts that ALL fail validation, reject the candidate outright instead of staging it for a review the user opted out of (or disguising it as a prose-only skill). The rejection is logged and recorded as a SEL tool-invocation event, and the staging condition drops the now-dead scripts_supplied disjunct (the reject branch consumes its only deciding case). The owning spec bullet in memory-skills-hooks.md is updated in the same commit to name the third outcome. Original change by Ruben Cuevas Menendez (rubencu). Rebased onto current main: the dashboard-handler hot-reload hunk was dropped because main now hot-reloads skills.* into the running HistoryConsolidator through its own live.watch_object subscription (reconfigure()), and the accompanying PATCH regression test was removed as subsumed by test_services_config_hot_reload.py. Co-authored-by: Kiro Crew <noreply@kirodotdev.github.io>
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
A prose-only update to an existing auto-skill can now be applied without review when the operator opts in with skills.auto_apply_updates on top of approval_required=false; disabling approval alone still stages every update. An update whose supplied scripts all failed validation follows the kirodotdev#7617 rule: rejected when approval is off, staged as prose when it is on. Every promotion, dismissal and unattended apply first captures the pending generation through authenticated no-follow reads, records its digest in a held claim lock, renames the public directory into a hidden quarantine and publishes only from a fresh private materialization of the captured bytes. Updates publish the complete live tree with no-replace renames and keep the replaced generation as permanent public evidence; restart recovery consumes, re-queues or retains each claim by exact digest and native identity. The private authority lives at tag-grants/auto-skill-private, beneath a directory every Kiro Crew sandbox already masked, so no new deny leaf, mask or spawn check is added and an authority problem never refuses an agent spawn. Gateway startup provisions it; a refusal only disables auto-skill staging and promotion. Hosts whose agents run in a sandbox Kiro Crew does not build never get an authority (fail-closed default). All auto-skill locks follow one documented order, skills.AUTO_SKILL_LOCK_ORDER: slug claim, claim, target, namespace. Fixes kirodotdev#3795
Problem / Motivation
With
skills.approval_required=false, a generated auto-skill candidate whose supplied scripts ALL fail static validation was still staged for review. The user had opted out of approval requests, yet a New skill awaiting review notification arrived anyway — for a candidate whose executable content had been entirely discarded, saved as if it were a prose-only skill.Why it matters
The Settings UI says approval is off while approval requests keep coming. Worse, the staged candidate hides what it was: it attempted to ship executable content, every script was rejected by the validator, and it now sits in the queue disguised as harmless prose. The notification even offers to disable a setting that is already disabled.
What changed (motivation → approach → change)
An all-invalid-scripts candidate is now rejected outright when approval is disabled. The staging decision in
HistoryConsolidator._process_auto_skillsgains a reject branch: a candidate that supplied scripts, kept none through validation, and has no review queue to land in (approval off) is dropped as a whole — logged and SEL-audited with reasonall_scripts_rejected— instead of being staged against the user's opt-out or auto-published. The stagingelifreadsself._approval_required or valid_scripts: the reject branch consumes the one case where a barescripts_suppliedused to decide, so that disjunct is gone.This fix deliberately covers NEW candidates only. UPDATE candidates for live skills always stage for human review regardless of
approval_required— an update overwrites a live skill, andapprove_pending_updateis the only path to live — so the opt-out does not apply there and that path is unchanged.Both prior safety boundaries hold. A candidate that retains any valid script still stages (only the passed scripts are kept). An all-invalid candidate still stages when approval is enabled, so a human can inspect the prose. A script-bearing candidate still never auto-publishes live. The owning spec bullet (
docs/system-specs/modules/memory-skills-hooks.md, "Staged approval") is updated in the same commit to name this third outcome.Tests
test_all_invalid_scripts_are_rejected_when_approval_disabled: an all-invalid-scripts candidate with approval off produces no live skill and no pending review entry.TestProcessAutoSkillsIntegrationkeeps proving that with approval enabled the prose still stages and the dangerous script is dropped.prove.pyPROVEN).Manual verification
N/A — unit coverage exercises script validation, live-skill creation, and pending-queue behavior directly.
Related Issues
No linked issue: reported directly from observed dashboard behavior.
Pattern harvest
Rule candidate: review-prompt
Pattern: a candidate whose executable payload was entirely rejected must not be re-classified as benign prose; the rejection must follow the candidate, not the payload.
Checklist
feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)docs/system-specs/modules/memory-skills-hooks.mdnames the new rejection outcome)