Unblocks self-review on this forge: setup-node is not mirrored (404); pip --user hits PEP 668. Optional setup-node + require runner Node; install PyYAML via workspace venv.
Unblocks self-review on this forge: setup-node is not mirrored (404); pip --user hits PEP 668. Optional setup-node + require runner Node; install PyYAML via workspace venv.
Gitea dogfood failed because actions/setup-node is not mirrored
(404 on git.potatolab.uk/actions/setup-node) and pip --user is
blocked by PEP 668. Prefer runner Node with an optional mirrored
setup-node, and install PyYAML into a workspace venv.
Sorted by severity. Each one spells out severity, where, and why it matters.
Fix or dismiss with dismissed: <reason> (then resolve the thread). Gatekeep has no chat / @-command surface.
Nothing the models agreed on — nice, or spooky. Glance at disputed below.
Disputed (summary only — not posted inline)
One model saw it; others didn't. Useful signal, not a merge blocker by itself.
No disagreements worth parking.
Deterministic security scan
No hard-fail hits.
Policy floor signals
none
Still on humans
Ensure .gatekeep/.venv is ignored by git to keep the repo clean.
Missed something real? Open a gatekeep-incident issue — that's how it becomes an eval. Inline comments only when models agree. PR tier is fail-closed. Findings lead with severity + location + why.
<!-- gatekeep-review -->
## Gatekeep review
Hey — here's the multi-model pass. **Severity first** so you can skim Critical → Low in seconds.
| | |
| --- | --- |
| **PR risk tier** | `high` |
| Policy floor | `low` |
| Agreement bar | ≥2 models |
| Agreed / disputed | 0 / 0 |
| Critical / High (agreed) | 0 / 0 |
Needs **2 human** approvals. Bots never fill those seats.
### Models (parallel)
| Model | Their PR tier | Findings |
| --- | --- | --- |
| `model-1-openrouter_google_gemma-4-31b-it_free.json` | `high` | 0 (err: http) |
| `model-2-openrouter_google_gemma-4-26b-a4b-it_free.json` | `high` | 0 (err: http) |
| `openrouter:cohere/north-mini-code:free` | `low` | 0 |
### Agreed findings
Sorted by severity. Each one spells out **severity**, **where**, and **why it matters**.
Fix or dismiss with `dismissed: <reason>` (then resolve the thread). Gatekeep has **no** chat / @-command surface.
_Nothing the models agreed on — nice, or spooky. Glance at disputed below._
### Disputed (summary only — not posted inline)
One model saw it; others didn't. Useful signal, not a merge blocker by itself.
_No disagreements worth parking._
### Deterministic security scan
_No hard-fail hits._
### Policy floor signals
- none
### Still on humans
- Ensure .gatekeep/.venv is ignored by git to keep the repo clean.
---
Missed something real? Open a **gatekeep-incident** issue — that's how it becomes an eval.
_Inline comments only when models agree. PR tier is fail-closed. Findings lead with severity + location + why._
Remove the optional setup-node step (still 404s as ##[error] on this
forge). Fix review.sh output writer — escaped quotes in a one-liner
caused SyntaxError after an otherwise successful publish.
layue13
merged commit a4fcfee1b3 into main2026-10-27 08:06:31 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Unblocks self-review on this forge: setup-node is not mirrored (404); pip --user hits PEP 668. Optional setup-node + require runner Node; install PyYAML via workspace venv.
Gatekeep review
Hey — here's the multi-model pass. Severity first so you can skim Critical → Low in seconds.
highlowNeeds 2 human approvals. Bots never fill those seats.
Models (parallel)
model-1-openrouter_google_gemma-4-31b-it_free.jsonhighmodel-2-openrouter_google_gemma-4-26b-a4b-it_free.jsonhighopenrouter:cohere/north-mini-code:freelowAgreed findings
Sorted by severity. Each one spells out severity, where, and why it matters.
Fix or dismiss with
dismissed: <reason>(then resolve the thread). Gatekeep has no chat / @-command surface.Nothing the models agreed on — nice, or spooky. Glance at disputed below.
Disputed (summary only — not posted inline)
One model saw it; others didn't. Useful signal, not a merge blocker by itself.
No disagreements worth parking.
Deterministic security scan
No hard-fail hits.
Policy floor signals
Still on humans
Missed something real? Open a gatekeep-incident issue — that's how it becomes an eval.
Inline comments only when models agree. PR tier is fail-closed. Findings lead with severity + location + why.