Repository navigation
fix(acp): backport skill-view visibility and bounds to release/0.8.0 - #15151
Closed
bolichen97 wants to merge 2 commits into
Closed
bolichen97 wants to merge 2 commits into
bolichen97 wants to merge 2 commits into
Conversation
This was referenced Sep 29, 2026
bolichen97
force-pushed
the
backport/skill-view-release-0.8.0
branch
from
September 29, 2026 23:40
30488f4 to
f1afbfd
Compare
Backport of kirodotdev#15132 (name skill-view aliases without MCP env values, keep the source path in the name) and kirodotdev#15125 (set_mode keeps the spawn alias, retries a miss and falls back to the authored agent) to release/0.8.0, squashed to keep the backport within two commits. (cherry picked from commit d6d0309) (cherry picked from commit 6313e50) (cherry picked from commit 5608475) Co-authored-by: buluoray <38930179+buluoray@users.noreply.github.com>
kiro-cli reloads ~/.kiro/agents only on a create, data write or remove of a *.json, never on the rename atomic_write publishes with. A new alias stays "Mode ... not found" until set_mode falls back to the authored agent, and an alias republished over a known name keeps its old content. A launcher that rewrites every spec on each launch also forced a rewrite of every alias per preparation and stranded ownership sidecars. - every publication ends in a same-bytes in-place rewrite, which the watcher reloads on and a racing reader never sees torn - currency compares parsed specs, so re-serialization is no rewrite - sidecars also bind the view identity; a moved-bytes pair is removed together, and orphaned sidecars and <alias>.lock residue are swept in bounded batches at boot and per spawn - a kirocrew doctor "skill-view residue" line Adds an inotify-driven harness that replays the incident over 50 spawns under a rewriting launcher. Fixes kirodotdev#15119 Refs kirodotdev#15121 (cherry picked from commit 7f60c56)
bolichen97
force-pushed
the
backport/skill-view-release-0.8.0
branch
from
September 30, 2026 00:11
f1afbfd to
e44ace4
Compare
This was referenced Sep 30, 2026
Contributor
Author
|
Posted on behalf of @bolichen97. Superseded by #15295, opened from a branch on kirodotdev/KiroCrew: release-branch PRs opened from a fork never start the review lanes. It carries the same backport, restacked on #15125 |
5 tasks done
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem / Motivation
Goal: Ship the 0.8.0.2 skill-view incident fix (#15121, #15119) on the 0.8 line, so a 0.8.0.x hot patch can turn native skill projection back on for hosts where another program rewrites
~/.kiro/agents.This is a backport of #15150 and the two PRs it stacks on (#15132, #15125) to
release/0.8.0. All three commits cherry-pick cleanly, andskill_projection.pyis byte-identical betweenrelease/0.8.0andmain. The root cause, the design and the evidence are in #15150.Why it matters
0.8.0.2 is on the stable channel. Every affected user either loses every turn (
Agent spec 'kirocrew-skill-view-...' is not installed) or runs withKIROCREW_NATIVE_SKILL_PROJECTION=0, which turns bounded skill discovery off.Not a goal
main.What changed (motivation → approach → change)
Two commits, following the release-branch backport shape of #11339 and #11645 (one combined PR, source SHAs as
cherry picked fromtrailers):mainfix(acp): backport skill-view alias naming and set_mode fallbackenvvalues and with the source path (skill_projection.py);set_modekeeps the spawn alias, retries a miss for 3 s, then runs the authored agent (acp/runtime.py,acp/session_handle.py), with thekirocrew.acp.skill_view.fallbackcounter (metrics/events.py,metrics.md)fix(acp): make skill-view publication visible and bounded.locksweeps (skill_projection.py), thekirocrew doctorresidue line (doctor_checks/resources.py,agent_sdk/drivers/acp.py), and the relaunch harnessEvery cherry-pick applied without conflict resolution. The spec text in
acp-client.mdcomes along with them. The design and evidence are in #15125 and #15150.Backwards compatibility
Compatible, for the same reasons as #15125 and #15150: alias names for an unchanged view and the existing sidecar keys are unchanged,
translatedefaults to the current send, and the only behaviour changes are on paths that failed or leaked before.Tests
The same tests as the source PRs, run against this branch's tree. The relaunch harness (
test/test_skill_view_relaunch_harness.py, 50 spawns × 4 agents under a rewriting launcher), the fallback suite (test/test_acp_set_mode_projection_fallback.py),test/test_native_skill_projection.py,test/test_cli_doctor.py,test/test_worker_agent.py,test/test_acp_runtime.pyandtest/test_acp_client.pyall pass (1,737 tests).Manual verification
As in #15150, measured on kiro-cli 2.26.0 and 2.25.0 in a throwaway
HOME. The 0.8.0.2 host behaviour this targets is the one reproduced there.Related Issues
Backport of #15132, #15125 and #15150. Refs #15121, #15119.
Pattern harvest
Not generalizable: a backport. The patterns are recorded in #15125 and #15150.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)acp-client.md,metrics.md, carried by the cherry-picks)