Creatidy/creatidy-kernel
5
0
Fork 0
Code Issues 8 Pull requests 1 Projects Releases Packages Wiki Activity Actions

[K5 correction] Create exact-head PRs through Forgejo AGit #25

Merged
adrian.tkacz merged 2 commits from issue-24-agit-pr into develop 2026-09-27 15:41:02 +00:00
Conversation 3 Commits 2 Files changed 8 +956 -717
kilo commented 2026-09-26 22:09:57 +00:00
Member
Copy link

Implements #24 as a separate correction after foundation PR #22 was integrated into develop; probe PR #23 remains DO NOT MERGE evidence only.

  • Replace snapshot-branch PR creation with exact full-SHA AGit push to refs/for/<base> using a fresh durable Operation topic push option; no ordinary mutable source branch for the PR effect.
  • Require a new-PR domain receipt and authoritative PR read-back of repository, base ref, identity and current full head; preserve observed base SHA as provenance and fail closed on collisions, uncertain delivery or moved head.
  • Update shared fake/reference conformance, transport tests and public Forge contract. No merge/deploy effect or core AGit payload.

Exact final HEAD 723fcfa9f5748c5629a486960ee17ea2c116daf8: make check passed (490 tests; lock/format/lint/basedpyright/imports/secrets/diff), make package-check passed; Forgejo CI run #2003 succeeded. A fresh verified independent full-PR review at this HEAD returned substantive APPROVE and zero blockers: #25 (comment) . Forgejo persisted COMMENT because the shared kilo account authored the PR; no formal platform approval is claimed. Delivery ledger and 1/10 remediation count: #24 (comment) . No second live AGit probe was run, as prohibited by #24. READY_FOR_OWNER_MERGE; do not auto-merge or close #24 until owner integration.

Implements #24 as a separate correction after foundation PR #22 was integrated into `develop`; probe PR #23 remains DO NOT MERGE evidence only. - Replace snapshot-branch PR creation with exact full-SHA AGit push to `refs/for/<base>` using a fresh durable Operation topic push option; no ordinary mutable source branch for the PR effect. - Require a new-PR domain receipt and authoritative PR read-back of repository, base ref, identity and current full head; preserve observed base SHA as provenance and fail closed on collisions, uncertain delivery or moved head. - Update shared fake/reference conformance, transport tests and public Forge contract. No merge/deploy effect or core AGit payload. Exact final HEAD `723fcfa9f5748c5629a486960ee17ea2c116daf8`: `make check` passed (490 tests; lock/format/lint/basedpyright/imports/secrets/diff), `make package-check` passed; Forgejo CI run #2003 succeeded. A fresh verified independent full-PR review at this HEAD returned substantive `APPROVE` and zero blockers: https://forgejo.creatidy.com/Creatidy/creatidy-kernel/pulls/25#issuecomment-12095 . Forgejo persisted `COMMENT` because the shared `kilo` account authored the PR; no formal platform approval is claimed. Delivery ledger and 1/10 remediation count: https://forgejo.creatidy.com/Creatidy/creatidy-kernel/issues/24#issuecomment-12096 . No second live AGit probe was run, as prohibited by #24. READY_FOR_OWNER_MERGE; do not auto-merge or close #24 until owner integration.
fix: create exact-head Forgejo PRs through AGit
All checks were successful
ci / check (pull_request) Successful in 28s
Details
9663fe754b
kilo reviewed 2026-09-26 22:18:48 +00:00
kilo left a comment
Copy link

PR Review: [K5 correction] Create exact-head PRs through Forgejo AGit (#25)

Repository: Creatidy/creatidy-kernel
Branch: issue-24-agit-pr -> develop
Frozen head: 9663fe754b
Linked issue: #24

Summary

The exact-SHA AGit creation shape, authoritative repository/base/head read-back, base-tip provenance, no-blind-retry guard, and adapter boundary are substantially present. Two issue acceptance requirements remain unsatisfied at this frozen head: producer-shaped nonzero AGit rejections are not classified, and later head movement is surfaced as an exception rather than stale evidence.

Intent / Acceptance Evidence

Issue #24 criterion Evidence at frozen head Result
Exact full SHA to refs/for/<base>; no ordinary branch forgejo_transport.py:399-460 constructs the exact-SHA refspec Satisfied
New-PR receipt plus authoritative repository/base/head read-back forgejo_transport.py:473-488, forgejo.py:373-391 Satisfied structurally
Base SHA is provenance, not authority core/forge.py:111-116 and read-back receipt Satisfied
Lost/ambiguous response is not blindly retried forgejo.py:303-304,357-375 Satisfied
Collision/update, permission, and unsupported outcomes remain distinct Nonzero porcelain is not parsed; see blocker 1 Not satisfied
Later moved head makes exact-head evidence stale Reconciliation raises ForgeConflict; see blocker 2 Not satisfied
No merge/deploy effect; dependency direction preserved No merge capability and AGit payload remains adapter-local Satisfied
Shared synthetic conformance Present, but it encodes the stale exception and lacks producer-shaped nonzero porcelain Partial
Required validation Forgejo run #2002 reports success at this SHA; PR reports make check (486 tests) and make package-check Satisfied, with job-detail endpoint unavailable

Architecture / Information Flow

The change keeps AGit-specific transport details in adapters and does not introduce a core dependency on Forgejo. Exact candidate and repository/base authority are read back before acknowledgement. The blockers concern domain-status normalization at the adapter boundary, not dependency direction.

Blocking Issues

  1. Nonzero AGit porcelain rejection evidence is discarded. In src/creatidy_kernel/adapters/forgejo_transport.py:467-472, any nonzero git push --porcelain result is classified only from stderr, then returns UNKNOWN before parsing stdout. Git reports proc-receive failures through porcelain status lines such as [remote rejected] (...); same-topic/non-fast-forward updates, unsupported AGit, and some permission failures can therefore collapse into UNKNOWN rather than the distinct rejected/collision/unsupported outcomes required by issue #24. tests/test_forge_transports.py:317-329 uses stderr-only fixtures and does not cover producer-shaped nonzero porcelain.

  2. A later moved PR head is not returned as stale evidence. src/creatidy_kernel/adapters/forgejo.py:379-388 (and the fake equivalent) raises ForgeConflict when authoritative read-back finds a changed head. tests/test_forge.py:315-339 explicitly expects that exception. Issue #24 requires later head movement to make recorded exact-head evidence stale, and EffectStatus.STALE exists for that domain distinction. Return a stale receipt for a previously accepted PR whose current head differs; retain fail-closed behavior during initial acknowledgement.

Non-blocking Suggestions

None.

Questions for the Author

None.

Security / Privacy Notes

Fail-closed handling is preserved, but collapsing known domain rejection evidence into UNKNOWN weakens the operation journal's ability to distinguish safe terminal refusal from ambiguous delivery.

Test / Documentation Gaps

tests/test_forge.py:404-410 verifies random topic uniqueness only. It does not exercise durable topic-to-Operation binding or rejection of reuse after a completed/closed topic, so the shared conformance suite does not fully demonstrate the documented never-reused topic responsibility.

Follow-Up Candidates

None. These gaps are required by issue #24 and belong in this PR.

Overall Recommendation

REQUEST_CHANGES

Blocking findings: 2. Re-review must bind to the remediated exact head.

Forgejo persistence note: the configured kilo identity authored this PR, so Forgejo rejected formal REQUEST_CHANGES with reject your own pull is not allowed. This review is therefore persisted as COMMENT; that state is not a formal approval or change-request review, while the substantive recommendation remains REQUEST_CHANGES.

## PR Review: [K5 correction] Create exact-head PRs through Forgejo AGit (#25) **Repository:** Creatidy/creatidy-kernel **Branch:** issue-24-agit-pr -> develop **Frozen head:** 9663fe754b9cc2ae1f602c9b1e8535a7be23abef **Linked issue:** #24 ### Summary The exact-SHA AGit creation shape, authoritative repository/base/head read-back, base-tip provenance, no-blind-retry guard, and adapter boundary are substantially present. Two issue acceptance requirements remain unsatisfied at this frozen head: producer-shaped nonzero AGit rejections are not classified, and later head movement is surfaced as an exception rather than stale evidence. ### Intent / Acceptance Evidence | Issue #24 criterion | Evidence at frozen head | Result | |---|---|---| | Exact full SHA to `refs/for/<base>`; no ordinary branch | `forgejo_transport.py:399-460` constructs the exact-SHA refspec | Satisfied | | New-PR receipt plus authoritative repository/base/head read-back | `forgejo_transport.py:473-488`, `forgejo.py:373-391` | Satisfied structurally | | Base SHA is provenance, not authority | `core/forge.py:111-116` and read-back receipt | Satisfied | | Lost/ambiguous response is not blindly retried | `forgejo.py:303-304,357-375` | Satisfied | | Collision/update, permission, and unsupported outcomes remain distinct | Nonzero porcelain is not parsed; see blocker 1 | Not satisfied | | Later moved head makes exact-head evidence stale | Reconciliation raises `ForgeConflict`; see blocker 2 | Not satisfied | | No merge/deploy effect; dependency direction preserved | No merge capability and AGit payload remains adapter-local | Satisfied | | Shared synthetic conformance | Present, but it encodes the stale exception and lacks producer-shaped nonzero porcelain | Partial | | Required validation | Forgejo run #2002 reports success at this SHA; PR reports `make check` (486 tests) and `make package-check` | Satisfied, with job-detail endpoint unavailable | ### Architecture / Information Flow The change keeps AGit-specific transport details in adapters and does not introduce a core dependency on Forgejo. Exact candidate and repository/base authority are read back before acknowledgement. The blockers concern domain-status normalization at the adapter boundary, not dependency direction. ### Blocking Issues 1. **Nonzero AGit porcelain rejection evidence is discarded.** In `src/creatidy_kernel/adapters/forgejo_transport.py:467-472`, any nonzero `git push --porcelain` result is classified only from `stderr`, then returns `UNKNOWN` before parsing `stdout`. Git reports proc-receive failures through porcelain status lines such as `[remote rejected] (...)`; same-topic/non-fast-forward updates, unsupported AGit, and some permission failures can therefore collapse into `UNKNOWN` rather than the distinct rejected/collision/unsupported outcomes required by issue #24. `tests/test_forge_transports.py:317-329` uses stderr-only fixtures and does not cover producer-shaped nonzero porcelain. 2. **A later moved PR head is not returned as stale evidence.** `src/creatidy_kernel/adapters/forgejo.py:379-388` (and the fake equivalent) raises `ForgeConflict` when authoritative read-back finds a changed head. `tests/test_forge.py:315-339` explicitly expects that exception. Issue #24 requires later head movement to make recorded exact-head evidence stale, and `EffectStatus.STALE` exists for that domain distinction. Return a stale receipt for a previously accepted PR whose current head differs; retain fail-closed behavior during initial acknowledgement. ### Non-blocking Suggestions None. ### Questions for the Author None. ### Security / Privacy Notes Fail-closed handling is preserved, but collapsing known domain rejection evidence into `UNKNOWN` weakens the operation journal's ability to distinguish safe terminal refusal from ambiguous delivery. ### Test / Documentation Gaps `tests/test_forge.py:404-410` verifies random topic uniqueness only. It does not exercise durable topic-to-Operation binding or rejection of reuse after a completed/closed topic, so the shared conformance suite does not fully demonstrate the documented never-reused topic responsibility. ### Follow-Up Candidates None. These gaps are required by issue #24 and belong in this PR. ### Overall Recommendation REQUEST_CHANGES Blocking findings: 2. Re-review must bind to the remediated exact head. Forgejo persistence note: the configured `kilo` identity authored this PR, so Forgejo rejected formal `REQUEST_CHANGES` with `reject your own pull is not allowed`. This review is therefore persisted as `COMMENT`; that state is not a formal approval or change-request review, while the substantive recommendation remains REQUEST_CHANGES.
fix: classify AGit rejection and stale PR evidence
All checks were successful
ci / check (pull_request) Successful in 29s
Details
723fcfa9f5
kilo reviewed 2026-09-26 22:28:38 +00:00
kilo left a comment
Copy link

PR Review: [K5 correction] Create exact-head PRs through Forgejo AGit (#25)

Repository: Creatidy/creatidy-kernel
Branch: issue-24-agit-pr -> develop
Frozen head: 723fcfa9f5
Linked issue: #24

Summary

The whole correction PR satisfies issue #24 at this exact head. It replaces snapshot-branch PR creation with an exact full-SHA AGit push, validates a new-PR domain receipt, authoritatively reads back repository/base/head identity, preserves base-tip provenance, fails closed on uncertainty, and exposes later head movement as stale evidence. The two blockers from the stale 9663fe7 review are remediated.

Intent / Acceptance Evidence

Issue #24 criterion Evidence at frozen head Result
Exact SHA to refs/for/<base>; no ordinary mutable source branch forgejo_transport.py:380-460; transport fixture asserts the full-SHA refspec and absence of HEAD/force options Satisfied
New-PR receipt plus authoritative repository/base/head read-back forgejo_transport.py:464-500; forgejo.py:343-395; shared repository-identity and mismatch tests Satisfied
Base SHA is provenance, not authority core/forge.py:111-116; moved-base conformance retains the observed base Satisfied
No blind retry after lost/ambiguous delivery forgejo.py:303-304,360-378; recovery requires a durable PR reference; tests retain the reference after temporary read-back loss Satisfied
Collision/update, permission, unsupported, and domain rejection remain distinct forgejo_transport.py:467-499; producer-shaped nonzero porcelain tests cover each reason Satisfied
Later moved head invalidates exact-head evidence forgejo.py:389-390 and fake equivalent return STALE with the durable PR reference Satisfied
Fresh never-reused Operation topic and topic collision refusal forge_refs.py:22-24,124-125; shared tests cover random topics, same-topic update refusal, and completed-topic reuse refusal Satisfied
No merge/deploy effect; dependencies remain inward AGit details remain adapter-local; Forge capabilities add no merge/deploy operation Satisfied
Validation and exact-head CI Issue ledger records make check (490 tests) and make package-check; Forgejo run #2003 succeeded at this SHA Satisfied

Architecture / Information Flow

AGit payload and porcelain parsing remain at the adapter edge. Core retains provider-neutral Effect, Receipt, and EffectStatus values. Acceptance requires both transport-domain creation evidence and authoritative HTTP read-back, so neither a zero Git exit nor an uploaded object alone grants acknowledgement.

Blocking Issues

None.

Non-blocking Suggestions

None.

Questions for the Author

None.

Security / Privacy Notes

The controlled Git environment, bounded output/timeout, exact object validation, disabled ambient Git configuration/hooks/credential helpers, and fail-closed parsing remain intact. No secrets, private payloads, merge authority, or deployment authority are introduced.

Test / Documentation Gaps

No blocking gap. A second live AGit probe was intentionally not run because issue #24 explicitly prohibits it; the README accurately describes the coverage as synthetic adapter conformance rather than a live integration claim.

Follow-Up Candidates

None.

Overall Recommendation

APPROVE

Blocking findings: none. This recommendation applies only to frozen HEAD 723fcfa9f5748c5629a486960ee17ea2c116daf8.

Forgejo persistence note: the configured kilo identity authored this PR, so Forgejo rejected formal APPROVED with approve your own pull is not allowed. This review is therefore persisted as COMMENT; that state is not a formal approval, while the substantive recommendation remains APPROVE.

## PR Review: [K5 correction] Create exact-head PRs through Forgejo AGit (#25) **Repository:** Creatidy/creatidy-kernel **Branch:** issue-24-agit-pr -> develop **Frozen head:** 723fcfa9f5748c5629a486960ee17ea2c116daf8 **Linked issue:** #24 ### Summary The whole correction PR satisfies issue #24 at this exact head. It replaces snapshot-branch PR creation with an exact full-SHA AGit push, validates a new-PR domain receipt, authoritatively reads back repository/base/head identity, preserves base-tip provenance, fails closed on uncertainty, and exposes later head movement as stale evidence. The two blockers from the stale `9663fe7` review are remediated. ### Intent / Acceptance Evidence | Issue #24 criterion | Evidence at frozen head | Result | |---|---|---| | Exact SHA to `refs/for/<base>`; no ordinary mutable source branch | `forgejo_transport.py:380-460`; transport fixture asserts the full-SHA refspec and absence of `HEAD`/force options | Satisfied | | New-PR receipt plus authoritative repository/base/head read-back | `forgejo_transport.py:464-500`; `forgejo.py:343-395`; shared repository-identity and mismatch tests | Satisfied | | Base SHA is provenance, not authority | `core/forge.py:111-116`; moved-base conformance retains the observed base | Satisfied | | No blind retry after lost/ambiguous delivery | `forgejo.py:303-304,360-378`; recovery requires a durable PR reference; tests retain the reference after temporary read-back loss | Satisfied | | Collision/update, permission, unsupported, and domain rejection remain distinct | `forgejo_transport.py:467-499`; producer-shaped nonzero porcelain tests cover each reason | Satisfied | | Later moved head invalidates exact-head evidence | `forgejo.py:389-390` and fake equivalent return `STALE` with the durable PR reference | Satisfied | | Fresh never-reused Operation topic and topic collision refusal | `forge_refs.py:22-24,124-125`; shared tests cover random topics, same-topic update refusal, and completed-topic reuse refusal | Satisfied | | No merge/deploy effect; dependencies remain inward | AGit details remain adapter-local; Forge capabilities add no merge/deploy operation | Satisfied | | Validation and exact-head CI | Issue ledger records `make check` (490 tests) and `make package-check`; Forgejo run #2003 succeeded at this SHA | Satisfied | ### Architecture / Information Flow AGit payload and porcelain parsing remain at the adapter edge. Core retains provider-neutral `Effect`, `Receipt`, and `EffectStatus` values. Acceptance requires both transport-domain creation evidence and authoritative HTTP read-back, so neither a zero Git exit nor an uploaded object alone grants acknowledgement. ### Blocking Issues None. ### Non-blocking Suggestions None. ### Questions for the Author None. ### Security / Privacy Notes The controlled Git environment, bounded output/timeout, exact object validation, disabled ambient Git configuration/hooks/credential helpers, and fail-closed parsing remain intact. No secrets, private payloads, merge authority, or deployment authority are introduced. ### Test / Documentation Gaps No blocking gap. A second live AGit probe was intentionally not run because issue #24 explicitly prohibits it; the README accurately describes the coverage as synthetic adapter conformance rather than a live integration claim. ### Follow-Up Candidates None. ### Overall Recommendation APPROVE Blocking findings: none. This recommendation applies only to frozen HEAD `723fcfa9f5748c5629a486960ee17ea2c116daf8`. Forgejo persistence note: the configured `kilo` identity authored this PR, so Forgejo rejected formal `APPROVED` with `approve your own pull is not allowed`. This review is therefore persisted as `COMMENT`; that state is not a formal approval, while the substantive recommendation remains APPROVE.
adrian.tkacz deleted branch issue-24-agit-pr 2026-09-27 15:41:02 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
adrian.tkacz
Labels
Clear labels   
bug
Something is not working

  
duplicate
This issue or pull request already exists

  
enhancement
New feature

  
help wanted
Need some help

  
invalid
Something is wrong

  
question
More information is needed

  
wontfix
This won't be fixed

No labels
bug
duplicate
enhancement
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
Creatidy/creatidy-kernel!25
No description provided.