Skip to content

Refresh the risk vocabulary from upstream - #400

Open
github-actions[bot] wants to merge 1 commit into
mainfrom
risk-vocab/2026-11-01
Open

github-actions[bot] wants to merge 1 commit into
mainfrom
risk-vocab/2026-11-01

Conversation

@github-actions

@github-actions github-actions Bot commented Nov 1, 2026

Copy link
Copy Markdown
Contributor

Monthly refresh of the vendored vocabulary. Read the numbers before merging.

Held out, after

split: 14727 train, 3901 held out (by path, not by row)
positive-class weights: destructive=6.1  sandbox_escape=3.7  credential_access=2.7  network=5.7  external_side_effect=9.0  safe=1.2

headline score on the holdout (max risky label, threshold 0.50):
  false alarms   168/2255 ordinary calls  (7.5%)
  missed          88/1646 risky calls     (5.3%)
  every false alarm:
    0.98  run shell command=rm -f dkms.conf
    0.98  run shell command=shred -u dkms.conf || rm -v dkms.conf
    0.98  run shell command=shred -u dkms.conf
    0.98  run shell command=strings web/bundles && rm dkms.conf || rm -v yarn-error.lo
    0.98  run shell command=rm dkms.conf; sed -n '1,40p' scripts/build.sh; tail -F dev

Held out, before

split: 14727 train, 3901 held out (by path, not by row)
positive-class weights: destructive=6.1  sandbox_escape=3.7  credential_access=2.7  network=5.7  external_side_effect=9.0  safe=1.2

headline score on the holdout (max risky label, threshold 0.50):
  false alarms   168/2255 ordinary calls  (7.5%)
  missed          88/1646 risky calls     (5.3%)
  every false alarm:
    0.98  run shell command=rm -f dkms.conf
    0.98  run shell command=shred -u dkms.conf || rm -v dkms.conf
    0.98  run shell command=shred -u dkms.conf
    0.98  run shell command=strings web/bundles && rm dkms.conf || rm -v yarn-error.lo
    0.98  run shell command=rm dkms.conf; sed -n '1,40p' scripts/build.sh; tail -F dev

The job refuses to open a PR at all if precision or recall dropped by more
than 0.05 on any label, so this is a proposal that already cleared that bar —
not a guarantee that the model got better.

scripts/fetch_risk_vocab.py re-fetched github/gitignore and the
gitleaks rules; the dataset was regenerated and the model retrained
from it. Held-out numbers are in the pull request body — they did not
regress, or this commit would not exist.

@claude claude Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Code review skipped — your organization has no extra usage available to pay for this review.

If your organization's extra usage balance is empty, an organization admin can add extra usage credits at claude.ai/admin-settings/usage. If its monthly spend limit was reached, an admin can raise it on the same page. If neither applies, contact Anthropic support.

Once extra usage is available, someone with write access to this repository can comment @claude review on this pull request to trigger a review.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0 participants