Conversation
…l-pr as one gated loop Adds .agent_loop/, a workflow-level orchestration package that runs the two review skills as a single loop: prefetch, one human break, then analysis fan-out, independent cross-model refutation, the seven review-pr gates, on-GPU validation, refutation of the executor's own findings, refold and cleanup. It is not a third skill, which is why it sits beside .claude rather than inside it: it calls the skills' CLI surface and adds the guarantees a single pass cannot give. Every card finding is attacked by an agent that never saw the analysis, that agent writes the ledger line itself, and the ledger's provenance is checked before a card can be certified. Six fail-closed guards, all covered by a zero-token dry-run harness (tools/dryrun.mjs, 10/10): dead analysis agent, collection short of the FIRE count, ledger provenance mismatch, gates never green, ledger rewritten behind the gate, and gates failing after the validation refold. docs/upstream-findings.md records four reproducible defects this loop found in validate-kernel-pr itself, each of which reads as a defect in the PR under review if the tool is trusted. Validated on #4961 across three heads on gfx950 and gfx1250; see VALIDATED-AGAINST.md for the pinned CLI surface and the measurements.
Contributor
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
PR title tags & labels: |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Comment on lines
+26
to
+27
| except Exception: | ||
| continue |
Contributor
| for line in raw.splitlines(): | ||
| try: | ||
| yield json.loads(line) | ||
| except Exception: |
Contributor
Comment on lines
+27
to
+28
| except Exception: | ||
| continue |
Contributor
| for line in raw.splitlines(): | ||
| try: | ||
| yield json.loads(line) | ||
| except Exception: |
Contributor
| @@ -0,0 +1,155 @@ | |||
| #!/usr/bin/env python | |||
Contributor
|
|
||
| runs = {} | ||
| rows = [] | ||
| for line in zstd.decompress(open(p, "rb").read()).decode("utf-8", errors="replace").splitlines(): |
Contributor
Comment on lines
+19
to
+20
| except Exception: | ||
| continue |
Contributor
| for line in zstd.decompress(open(p, "rb").read()).decode("utf-8", errors="replace").splitlines(): | ||
| try: | ||
| o = json.loads(line) | ||
| except Exception: |
Contributor
| rows.append((d.get("seq"), d.get("phase"), d.get("label"), d.get("childId"))) | ||
|
|
||
| for seq, ph, label, cid in rows: | ||
| print(f"seq {seq:>3} {str(ph):<26} {str(label):<44} {cid}") |
Contributor
| rows.append((d.get("seq"), d.get("phase"), d.get("label"), d.get("childId"))) | ||
|
|
||
| for seq, ph, label, cid in rows: | ||
| print(f"seq {seq:>3} {str(ph):<26} {str(label):<44} {cid}") |
Contributor
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
2 participants
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.
…l-pr as one gated loop
Adds .agent_loop/, a workflow-level orchestration package that runs the two review skills as a single loop: prefetch, one human break, then analysis fan-out, independent cross-model refutation, the seven review-pr gates, on-GPU validation, refutation of the executor's own findings, refold and cleanup.
It is not a third skill, which is why it sits beside .claude rather than inside it: it calls the skills' CLI surface and adds the guarantees a single pass cannot give. Every card finding is attacked by an agent that never saw the analysis, that agent writes the ledger line itself, and the ledger's provenance is checked before a card can be certified.
Six fail-closed guards, all covered by a zero-token dry-run harness (tools/dryrun.mjs, 10/10): dead analysis agent, collection short of the FIRE count, ledger provenance mismatch, gates never green, ledger rewritten behind the gate, and gates failing after the validation refold.
docs/upstream-findings.md records four reproducible defects this loop found in validate-kernel-pr itself, each of which reads as a defect in the PR under review if the tool is trusted.
Validated on #4961 across three heads on gfx950 and gfx1250; see VALIDATED-AGAINST.md for the pinned CLI surface and the measurements.