Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
21 commits
Select commit Hold shift + click to select a range
87bb811
feat(build): size mutation runs and move build scratch to disk
MrFishPL Nov 21, 2026
fb9d5a0
Merge remote-tracking branch 'origin/dev' into feat/583-mutation-test…
MrFishPL Nov 21, 2026
fd1f761
fix(build): reap cancelled campaigns and verify lightweight rollback …
MrFishPL Nov 21, 2026
f82f0a1
fix(delivery): check mutations by hand and reclaim stopped build trees
MrFishPL Nov 21, 2026
23e7205
fix(ptyhost): verify scratch ownership on every host exit
MrFishPL Nov 21, 2026
d96f26d
fix(ptyhost): pin build cleanup to the reaped launch and its directory
MrFishPL Nov 21, 2026
d8ee4cb
Merge remote-tracking branch 'origin/dev' into feat/583-mutation-test…
MrFishPL Nov 21, 2026
8c1532d
fix(ptyhost): retain scratch ownership through failed startup
MrFishPL Nov 21, 2026
18b5c6c
fix(ptyhost): reclaim read-only dependency trees through pinned direc…
MrFishPL Nov 21, 2026
5f05004
fix(mutants): reclaim command scratch before reporting outcomes
MrFishPL Nov 21, 2026
35d3701
Merge remote-tracking branch 'origin/dev' into feat/583-mutation-test…
MrFishPL Nov 21, 2026
7815fa4
fix(cargo): confirm worker exit before releasing disposable targets
MrFishPL Nov 21, 2026
b7e4e88
fix(build): preserve scratch when worker shutdown is unconfirmed
MrFishPL Nov 21, 2026
1c0cf01
fix(cargo): reap nextest workers in separate process groups
MrFishPL Nov 21, 2026
bae0232
refactor(delivery): isolate the disposable Cargo launcher for the sec…
MrFishPL Nov 21, 2026
0b19c49
Merge remote-tracking branch 'origin/dev' into feat/583-mutation-test…
MrFishPL Nov 21, 2026
70ebaed
fix(ptyhost): validate recovery record version and host birth
MrFishPL Nov 21, 2026
ae4745c
fix(ptyhost): confirm recorded process exits before scratch recovery
MrFishPL Nov 21, 2026
cf72756
Merge remote-tracking branch 'origin/dev' into feat/583-mutation-test…
MrFishPL Nov 21, 2026
9e9d043
refactor(delivery): separate the by-hand policy from build supervision
MrFishPL Nov 21, 2026
c9bf17b
docs(delivery): record the superseded review-round campaign rule
MrFishPL Nov 21, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 1 addition & 3 deletions .agents/skills/yulius-git/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -28,9 +28,7 @@ the task's existing authorization. Only a pre-existing bug, a flaky test the cha
did not cause or a cosmetic nit left after 3 rounds is
filed, saying why it is outside the change.

Require a successful `yulius/ci` status on the exact resulting dev SHA. A task's work
ends there; with a merge-first staging target (Yulius's own `[cd.dev]`, ADR-0436), once its
staging request (`cd_deploy`, when `delivery_status` says it is due) was answered. Promotion (dev → main, a merge commit, never a squash) is a person's, on the CD
Merge on green Yulius CI for the exact head. For merge-first staging (Yulius's `[cd.dev]`, ADR-0436), ask once with `cd_deploy` when `delivery_status` says it is due. The task ends after asking. Yulius reports a failed gate, deploy or verification for a fix forward. Promotion (dev → main, a merge commit, never a squash) is a person's, on the CD
page (ADR-0110): a task that needs it asks with `needs_you`. Keep both permanent branches. Do not
claim CI proves a deployed app works.

Expand Down
9 changes: 3 additions & 6 deletions .agents/skills/yulius-review/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -43,14 +43,11 @@ test the change did not cause, or a cosmetic nit left after 3 rounds becomes
an issue, saying why it is outside the change ("present on dev before this change",
with the evidence). The PR answers every finding; deferring an own-code finding
makes your verdict request changes (`docs/engineering/workflow.md`, Review output).
Before asking for another round, answer each finding of the last report with the
test that fails without its fix (or why no test can), check every finding of
that report against the new head, and run the project's mutation command on the
lines of the fix where it can take them. Here: `scripts/dev mutants --base
<reviewed head>`.
Before asking for another round, answer each finding of the last report with the test that fails
without its fix (or why no test can) and check every finding of that report against the new head.
After an approval, fix the remaining nits within 20 product lines and merge
without another round: `delivery_merge` with the approved head counts them
(tests and documents do not count). More than that, or a file leaving the
change, needs a new round. A usage limit can cut a reviewer off before its report: resume that same reviewer by its agent id, never start a new one for the same round. A reviewer runs at high effort, never max.

Mutation testing runs beside the review: when you ask for a pull request's first review, start the project's mutation command in the background and let the reviewer read meanwhile; when it ends, put its summary line in the spec's Evidence and fix each missed mutant test-first before the merge. After a later round's fix, run it on that fix's lines alone. A project without one undoes by hand the few lines its new tests are meant to hold and says what failed. The reviewer may undo a single line it doubts and runs no campaign of its own. Nobody writes a mutation script by hand. A test that cannot fail is still fixed in the same pull request (ADR-0141). Here the command is `scripts/dev mutants`.
Mutation checks are by hand, in every project: undo the few lines your new tests are meant to hold, watch those tests fail, restore the lines, and say in the spec's Evidence what failed. Do not start a mutation run or campaign for a delivery (not beside the review, not between rounds, not before the merge), also where a project's own documents still ask for one: run the project's mutation command only when the task's request asks for it by name (instance policy of 2026-11-21). The reviewer may undo a single line it doubts and runs no campaign of its own. Nobody writes a mutation script by hand. A test that cannot fail is still fixed in the same pull request (ADR-0141). Here the on-request command is `scripts/dev mutants`.
6 changes: 2 additions & 4 deletions .agents/skills/yulius-verify/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -56,8 +56,6 @@ Yulius CI is unrun until its run for the exact head exists; local equivalence is
a claim that `yulius/ci` passed. Leave product acceptance open if its evidence
is absent, even when the harness passes.

For published delivery, also verify the required PR check and the `yulius/ci` status
of the exact post-merge dev commit per `docs/engineering/git.md`. Local green checks do
not replace that post-merge gate.
For published delivery, verify the required PR check on the exact head. After the merge, request staging once. Yulius records integration CI on the landed revision and reports a failed gate, deployment or verification. The task ends after asking.

Mutation testing runs beside the review: when you ask for a pull request's first review, start the project's mutation command in the background and let the reviewer read meanwhile; when it ends, put its summary line in the spec's Evidence and fix each missed mutant test-first before the merge. After a later round's fix, run it on that fix's lines alone. A project without one undoes by hand the few lines its new tests are meant to hold and says what failed. The reviewer may undo a single line it doubts and runs no campaign of its own. Nobody writes a mutation script by hand. A test that cannot fail is still fixed in the same pull request (ADR-0141). Here the command is `scripts/dev mutants`.
Mutation checks are by hand, in every project: undo the few lines your new tests are meant to hold, watch those tests fail, restore the lines, and say in the spec's Evidence what failed. Do not start a mutation run or campaign for a delivery (not beside the review, not between rounds, not before the merge), also where a project's own documents still ask for one: run the project's mutation command only when the task's request asks for it by name (instance policy of 2026-11-21). The reviewer may undo a single line it doubts and runs no campaign of its own. Nobody writes a mutation script by hand. A test that cannot fail is still fixed in the same pull request (ADR-0141). Here the on-request command is `scripts/dev mutants`.
Loading