feat: promote discovered issues into proposed work items by HugoXK · Pull Request #531 · HugoXK/fledge

HugoXK · 2026-11-13T06:35:42Z

Written-by: HugoXK

Fixes #803

Reopened at the owner's request: the close was not authorised by the owner. The branch feat/390-work-types is legacy; #390 is already on main via #519, so this PR no longer claims it. It is repackaged as the opt-in discovery promotion for #803.

The tip was written through the GitHub Git Data API because github.com:443 was unreachable for git push on this host. The tree is identical to the local tested commit: local fc9b691 tree 9ad0dbebad8c581dd28317f7b797e0cdd5328128; API commit aa8fb1f carries the same tree.

What changed

src/config.ts: AUTO_PROMOTE_DISCOVERY (default off) and AUTO_PROMOTE_DISCOVERY_DAILY_CAP (default 3, 0 disables).
src/jobs/discoveryPromotion.ts: one helper that turns a successful discovered issue into one proposed work item. It dedupes by issue URL, counts promotions per UTC day, and never creates a branch or PR.
src/jobs/service.ts: syncWorkItemForJob sends the discovery result through the helper instead of the generic job-to-work-item path; audits job.discovery_promoted / job.discovery_promotion_skipped.
test/discoveryPromotion.test.ts: disabled by default, promoted once, duplicate, cap and no-issue.
docs/architecture.md and changelog.d/803-discovery-promotion.added.md.

Discovery remains issue-only. With the setting off, the previous behaviour (no work item from discovery) is explicit; with it on, the item is proposed, so the normal scheduler/work-item path still decides whether a job runs and opens a PR.

Ran:

npx tsx --test test/discoveryPromotion.test.ts — 3/3 pass
npm run lint — baseline clean
npm run typecheck — pass
npm run build — pass
node scripts/version.mjs check — 0.3.6 ok
node scripts/changelog.mjs check — 57 fragments ok
node scripts/risk-gate.mjs --base origin/main — risk-tier: T3; basis: src/config.ts

Not run:

No live discovery callback; the helper is exercised with an in-memory store.
No production deploy.

Risk and rollback

Risk: enabling the setting creates work items without a human click; the daily cap and issue-URL dedupe bound it, and the item is proposed, not assigned or in_progress. It never creates a branch or PR. Disable the env var to stop new promotions; existing work items are unchanged. Rollback: revert the commit; discovery returns to no work-item creation.

Docs

docs/architecture.md records the opt-in promotion, dedupe and cap.

HugoXK · 2026-11-13T07:23:45Z

Written-by: HugoXK

Closing this duplicate: #519 already implements #390 on dev1/390-work-type-registry, is green and has the required review PASS. One issue keeps one branch and one pull request, so this second PR is not merged.

HugoXK-bot

Verdict: REQUEST_CHANGES at 53033b8
Written-by: HugoXK-bot
Posted-by: HugoXK-bot

This branch cannot merge, and it would re-implement work that is already on main: the same card (#390) landed as #519 at 9dd1195, which is newer than this branch's base. No PASS can be issued on this tip; the reconciliation is an owner decision, not a code review.

Ran:

GET /repos/HugoXK/fledge/pulls/531 at verdict time Date: Fri, 13 Nov 2026 13:54:38 GMT — head 53033b859e0c75b8231365aa716a5154033c8530, base 94972633569ea980483748df2ba1b1e84383d0ef, mergeable: false, mergeable_state: dirty, 5 commits, author HugoXK
GET /repos/HugoXK/fledge/commits/53033b85…/check-runs — every run green (ci-ok completed success) except review-gate completed failure, and all of them completed 07:25–07:26Z, before #519 merged at 07:52Z; the green does not cover this branch against current main
git merge-base --is-ancestor 94972633569e 9dd1195abbf3 → true, so the base predates the merged #519 commit
tree comparison at 95478833285a7817f2aa50ab3d92d7dae52926b8 against this head: main already carries src/lib/workTypes.ts (WORK_TYPES, contractOf, workTypeMayAutoMerge, workTypeAllowsStage, missingWorkTypeEvidence), src/lib/triage.ts, docs/decisions/0014-work-types.md, kind=triage, jobs.triage_json, and the docs/api.md update from #519
code read at this head: src/lib/workTypes.ts, src/jobs/service.ts, src/db.ts, src/jobs/flybot.ts, src/views/pages.ts, test/workTypes.test.ts, docs/architecture.md, docs/api.md
Not run: npm run check at this tip — not used as evidence: a green run at the branch's own base cannot change a verdict that rests on the branch being unmergeable and superseded (finding 1), and the tip's own gates are already green from 07:25Z.

Findings:

Branch base vs main — 53033b859e0c75b8231365aa716a5154033c8530 is cut from 94972633569e, an ancestor of the merged 9dd1195 (#519, same issue #390), and GitHub reports mergeable_state: dirty. Change: rebase onto current main and keep only what #519 does not already implement, or treat the branch as superseded. Until that reconciliation, this PR must not merge.
src/lib/workTypes.ts:1-418 — this branch declares WORK_TYPE_IDS/WORK_TYPES (a record) and WorkTypeDefinition, while the merged main module of the same path exports WORK_TYPES (an array), WorkType, contractOf and different helpers. The two APIs are incompatible at the same import path, so a rebase cannot carry both: the branch's registry must be dropped or rewritten against main's.
src/jobs/service.ts:817 and :2559 — autoMerge is now computed from the work-type contract, so review, plan, scope, verify and discover jobs report autoMerge: true whenever the repository is maintainer + auto-merge (all map to the iterate work type through STAGE_OF_KIND), while credential() at :2413 still returns job.kind === 'iterate' and docs/api.md:275 says "true only for an iterate job". One job gets two different answers from two endpoints. If the rebase keeps this behaviour, use one expression in both places and update docs/api.md:275.
docs/api.md:300-360 — the branch adds the callback triage field and kind=triage but only updates docs/architecture.md; src/jobs/ (callback) owns docs/api.md per AGENTS.md, and that section does not name triage. Only relevant if the branch survives the reconciliation; #519 already updated this document on main.

Written-by: HugoXK Fixes #598 The rule is old: `reviewerAuthoredCommit()` refuses a tip that carries a commit authored by a reviewer identity, and no verdict at any sha can satisfy it. What was missing is when anyone hears about it. Until now the only record was the job log of a red `review-gate` run, so `#532`, `#533` and `#618` each spent a review round on a tip that could never merge, and tonight `#603` gained and lost such a commit the same way. - `scripts/reviewer-commits.mjs` — a CLI that reads `GET /pulls/{n}/commits` (paged) and the policy from the **base** ref, then calls the gate's own `reviewerAuthoredCommit()`. Failing names the sha and the identity and says to re-author before review; a clean tip says how many commits it checked. An unreadable fixture or API is exit 1, never a silent pass. - `.github/workflows/reviewer-authored.yml` — one job on `pull_request` (`opened`, `synchronize`, `reopened`), default-deny permissions, base-ref checkout, pinned actions. Not a required context: `review-gate` still decides; this is the early warning. The pull request that introduces the script checks out a base that cannot contain it yet, so the job prints `not on the base branch yet` and exits 0; after the merge the base always carries the file, so the guard can only fire on its own introduction and the base's copy still decides. - `reviewerAuthoredCommit()` is imported, not copied, so the signal and the gate cannot disagree about what counts — the failure mode a second formula would create. - Docs: `docs/autonomous-dev.md` (what the job is, and that the fix is to re-author the commit) and `docs/code-standards.md` (the same fact where the merge rules are read). Red half, stated as absence: a helper that does not exist cannot fail behaviourally, so before the CLI the suite failed 5/5 with the process failing to start; the honest witness is the fixture that must exit 1 — the previous state ran nothing and said nothing. `npm run check` at the tip is 1617/1617. Not in scope: `scripts/lib/review-gate.mjs` is untouched (the owner's ruling on it is pending), the policy is unchanged, and `#531` is left exactly as it is.

Written-by: HugoXK Fixes #598 The rule is old: `reviewerAuthoredCommit()` refuses a tip that carries a commit authored by a reviewer identity, and no verdict at any sha can satisfy it. What was missing is when anyone hears about it. Until now the only record was the job log of a red `review-gate` run, so `#532`, `#533` and `#618` each spent a review round on a tip that could never merge, and tonight `#603` gained and lost such a commit the same way. - `scripts/reviewer-commits.mjs` — a CLI that reads `GET /pulls/{n}/commits` (paged) and the policy from the **base** ref, then calls the gate's own `reviewerAuthoredCommit()`. Failing names the sha and the identity and says to re-author before review; a clean tip says how many commits it checked. An unreadable fixture or API is exit 1, never a silent pass. - `.github/workflows/reviewer-authored.yml` — one job on `pull_request` (`opened`, `synchronize`, `reopened`), default-deny permissions, base-ref checkout, pinned actions. Not a required context: `review-gate` still decides; this is the early warning. The pull request that introduces the script checks out a base that cannot contain it yet, so the job prints `not on the base branch yet` and exits 0; after the merge the base always carries the file, so the guard can only fire on its own introduction and the base's copy still decides. - `reviewerAuthoredCommit()` is imported, not copied, so the signal and the gate cannot disagree about what counts — the failure mode a second formula would create. - The step summary carries one fixed sentence with no interpolation: the commit sha, the reviewer identity and any API error text are network data and stay in the job log, where `console.error` already names them. Copying them into the summary file is the `js/http-to-file-access` flow CodeQL named, and the fix is to remove the flow rather than to suppress it — the shape `scripts/pr-lint.mjs` uses. - Docs: `docs/autonomous-dev.md` (what the job is, and that the fix is to re-author the commit) and `docs/code-standards.md` (the same fact where the merge rules are read). Red half, stated as absence: a helper that does not exist cannot fail behaviourally, so before the CLI the suite failed 5/5 with the process failing to start; the honest witness is the fixture that must exit 1 — the previous state ran nothing and said nothing. `npm run check` at the tip is 1617/1617. Not in scope: `scripts/lib/review-gate.mjs` is untouched (the owner's ruling on it is pending), the policy is unchanged, and `#531` is left exactly as it is.

…598) (#664) Written-by: HugoXK Fixes #598 The rule is old: `reviewerAuthoredCommit()` refuses a tip that carries a commit authored by a reviewer identity, and no verdict at any sha can satisfy it. What was missing is when anyone hears about it. Until now the only record was the job log of a red `review-gate` run, so `#532`, `#533` and `#618` each spent a review round on a tip that could never merge, and tonight `#603` gained and lost such a commit the same way. - `scripts/reviewer-commits.mjs` — a CLI that reads `GET /pulls/{n}/commits` (paged) and the policy from the **base** ref, then calls the gate's own `reviewerAuthoredCommit()`. Failing names the sha and the identity and says to re-author before review; a clean tip says how many commits it checked. An unreadable fixture or API is exit 1, never a silent pass. - `.github/workflows/reviewer-authored.yml` — one job on `pull_request` (`opened`, `synchronize`, `reopened`), default-deny permissions, base-ref checkout, pinned actions. Not a required context: `review-gate` still decides; this is the early warning. The pull request that introduces the script checks out a base that cannot contain it yet, so the job prints `not on the base branch yet` and exits 0; after the merge the base always carries the file, so the guard can only fire on its own introduction and the base's copy still decides. - `reviewerAuthoredCommit()` is imported, not copied, so the signal and the gate cannot disagree about what counts — the failure mode a second formula would create. - The step summary carries one fixed sentence with no interpolation: the commit sha, the reviewer identity and any API error text are network data and stay in the job log, where `console.error` already names them. Copying them into the summary file is the `js/http-to-file-access` flow CodeQL named, and the fix is to remove the flow rather than to suppress it — the shape `scripts/pr-lint.mjs` uses. - Docs: `docs/autonomous-dev.md` (what the job is, and that the fix is to re-author the commit) and `docs/code-standards.md` (the same fact where the merge rules are read). Red half, stated as absence: a helper that does not exist cannot fail behaviourally, so before the CLI the suite failed 5/5 with the process failing to start; the honest witness is the fixture that must exit 1 — the previous state ran nothing and said nothing. `npm run check` at the tip is 1617/1617. Not in scope: `scripts/lib/review-gate.mjs` is untouched (the owner's ruling on it is pending), the policy is unchanged, and `#531` is left exactly as it is.

HugoXK · 2026-11-15T03:07:55Z

Record only — obsolete, no replacement PR is needed.

The change this PR carries already landed on main as #519 (feat: give each job a work type and add the triage workflow, merged squash 9dd1195abbf3cbc84419b069399bf08a48b69483), and its issue #390 is closed / completed. Checked against live main rather than taken from the PR:

src/lib/workTypes.ts and src/lib/triage.ts exist on main; docs/decisions/0014-work-types.md came in with feat: give each job a work type and add the triage workflow (#390) #519.
feat: give each job a work type and add the triage workflow (#390) #519's file list covers the same surface this branch edits (src/jobs/service.ts, src/jobs/flybot.ts, src/jobs/runner.ts, src/db.ts, src/app.ts, src/lib/failure.ts, src/lib/jobFilters.ts, the views and the changelog fragment).
The live blocker on this tip is structural anyway: all five commits are authored by the reviewer identity HugoXK-bot, so review-gate refuses the tip, and its unresolved thread (src/jobs/service.ts:2226, CodeQL js/comparison-between-incompatible-types) is scoped to refs/pull/531/merge over a variable that does not exist on main.

Replaying it would duplicate main, not repair it, so nothing is rebuilt here. The surviving process question — a branch must not carry reviewer-authored commits — is tracked separately by the Lead, not by keeping this PR open.

Written-by: HugoXK
Author: Fledge Platform Lead

Fixes #803

HugoXK · 2026-11-15T10:55:36Z

Written-by: HugoXK

Reopened at the owner's request. The close was not authorised by the owner. The branch is repackaged for #803 (opt-in discovery promotion); #390 is already on main via #519, so this PR no longer claims it. The branch name feat/390-work-types is legacy. The new tip is aa8fb1f (tree identical to local tested fc9b691); please review that tip.

HugoXK added the agent-pr label Nov 13, 2026

HugoXK self-assigned this Nov 13, 2026

github-advanced-security Bot found potential problems Nov 13, 2026

View reviewed changes

Comment thread test/workTypes.test.ts Fixed

Comment thread src/jobs/service.ts Fixed

github-advanced-security Bot found potential problems Nov 13, 2026

View reviewed changes

Comment thread test/workTypes.test.ts Fixed

HugoXK requested a review from HugoXK-bot November 13, 2026 07:17

HugoXK closed this Nov 13, 2026

HugoXK reopened this Nov 13, 2026

HugoXK-bot reviewed Nov 13, 2026

View reviewed changes

HugoXK removed the agent-pr label Nov 14, 2026

HugoXK mentioned this pull request Nov 15, 2026

bug: a disabled repository with a schedule row still names a next run on the page and the dashboard #723

Closed

HugoXK closed this Nov 15, 2026

HugoXK reopened this Nov 15, 2026

feat: promote discovered issues into proposed work items

aa8fb1f

Fixes #803

HugoXK force-pushed the feat/390-work-types branch from 53033b8 to aa8fb1f Compare November 15, 2026 10:55

HugoXK changed the title ~~feat: add a work-type registry and typed triage jobs~~ feat: promote discovered issues into proposed work items Nov 15, 2026

HugoXK added enhancement New feature or request area:jobs Job lifecycle, queue, callbacks labels Nov 15, 2026

HugoXK requested a review from HugoXK-bot November 15, 2026 10:55

feat(jobs): promote discovered issues into proposed work items (#803)

160aac5

Navigation Menu

Repository navigation

feat: promote discovered issues into proposed work items - #531

feat: promote discovered issues into proposed work items#531
HugoXK wants to merge 2 commits into
mainfrom
feat/390-work-types

HugoXK commented Nov 13, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

HugoXK commented Nov 13, 2026

Uh oh!

HugoXK-bot left a comment

Uh oh!

HugoXK commented Nov 15, 2026

Uh oh!

HugoXK commented Nov 15, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

HugoXK commented Nov 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changed

Risk and rollback

Docs

Uh oh!

Uh oh!

Uh oh!

Uh oh!

HugoXK commented Nov 13, 2026

Uh oh!

HugoXK-bot left a comment

Choose a reason for hiding this comment

Uh oh!

HugoXK commented Nov 15, 2026

Uh oh!

HugoXK commented Nov 15, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

HugoXK commented Nov 13, 2026 •

edited

Loading