Skip to content

Add Model Optimizer, Megatron Core, and Megatron Bridge to catalog - #2

Merged
mosheabr merged 1 commit into
NVIDIA:mainfrom
shljessie:add-nvidia-repos-to-catalog
Apr 7, 2026
Merged

Add Model Optimizer, Megatron Core, and Megatron Bridge to catalog#2
mosheabr merged 1 commit into
NVIDIA:mainfrom
shljessie:add-nvidia-repos-to-catalog

Conversation

@shljessie

@shljessie shljessie commented Apr 6, 2026

Copy link
Copy Markdown
Contributor
  • Add three NVIDIA repos to both the Available Skills and Getting Help tables.
  • Update TensorRT-LLM skill count from 3 to 4.

Add three NVIDIA repos to both the Available Skills and Getting Help
tables. Update TensorRT-LLM skill count from 3 to 4. All links were
verified against live GitHub pages; repos without Discussions or
SECURITY.md are marked with an em dash.

Made-with: Cursor
@shljessie
shljessie requested a review from mosheabr as a code owner April 6, 2026 18:58
@mosheabr mosheabr self-assigned this Apr 6, 2026

@mosheabr mosheabr left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Review: Approve ✓

Skill counts — all verified against source repos:

  • TensorRT-LLM: 4 ✓ (was 3, serve-config-guide added upstream)
  • Model Optimizer: 4 ✓ (common, deployment, evaluation, ptq)
  • Megatron Core: 2 ✓ (build-and-test, respond-to-issue)
  • Megatron Bridge: 8 ✓ (8 skills confirmed)

Getting Help table links verified:

  • Model Optimizer: Discussions correct (disabled on repo). CONTRIBUTING.md and SECURITY.md exist ✓
  • Megatron-LM: SECURITY.md does not exist in repo — is correct ✓
  • Megatron-Bridge: SECURITY.md does not exist in repo — is correct ✓

Minor suggestion (non-blocking):

  • Consider adding a CHANGELOG.md entry (v0.3.0) for consistency with PR #1 which included one. Not a blocker.

LGTM — counts are accurate, links are valid.

@mosheabr
mosheabr merged commit 4e64ee0 into NVIDIA:main Apr 7, 2026
mosheabr added a commit that referenced this pull request Jun 4, 2026
#2 (MED) — gh issue create can skip tracker on bad assignee:
  The "Track dropped skills" step has continue-on-error: true, so a
  bad assignee wouldn't kill the whole sync — but it would skip
  tracker creation entirely, defeating the feature. Both branches
  (edit existing / create new) now follow the same tolerant pattern:
  create-without-assignees first, then loop --add-assignee in a
  trim + per-handle warning. A typo in CATALOG_TRACKER_ASSIGNEES no
  longer blocks the tracker from existing.

#3 (LOW) — Hardcoded personal fallback:
  Reworked per Codex's catch that GitHub issue assignees are users,
  not teams — so the earlier "use a team handle" comment was wrong.
  If CATALOG_TRACKER_ASSIGNEES is unset we now skip assignment
  entirely and emit a ::warning title=Drift notifications inactive::.
  No personal fallback; a deactivated handle would silently mute
  notifications. The warning surfaces in the workflow log every run
  until the secret is set.

#4 (LOW / clarification) — Deleted-sig case:
  Added an inline comment in the drift loop noting that an rsync
  that deletes skill.oms.sig alongside content edits registers the
  deletion in `git diff` (sig_changed matches → drift NOT flagged
  here), and the now-sigless skill falls through to the compliance
  step under "missing artifacts: skill.oms.sig". Defense in depth.

#1 (HIGH per Sayali) — README version inconsistency on revert:
  NOT addressed in this commit — verified against
  .github/scripts/regenerate-readme.sh on main: the README table
  is now "Product | Description | Skills" (PR #215, commit 1104de0
  on 2026-06-01 dropped the Source + Version columns). VERSIONS_FILE
  is declared but never consumed, and /tmp/sync-versions.txt is
  vestigial. So the exact "README will lie" risk Sayali raised is
  stale. Plan to send her the context separately and open a small
  follow-up PR to drop the dead sync-versions write + VERSIONS_FILE
  declaration so this trap doesn't catch a future reader.

Signed-off-by: Moshe Abramovitch <moshea@nvidia.com>
mosheabr added a commit that referenced this pull request Jun 8, 2026
#2 (MED) — gh issue create can skip tracker on bad assignee:
  The "Track dropped skills" step has continue-on-error: true, so a
  bad assignee wouldn't kill the whole sync — but it would skip
  tracker creation entirely, defeating the feature. Both branches
  (edit existing / create new) now follow the same tolerant pattern:
  create-without-assignees first, then loop --add-assignee in a
  trim + per-handle warning. A typo in CATALOG_TRACKER_ASSIGNEES no
  longer blocks the tracker from existing.

#3 (LOW) — Hardcoded personal fallback:
  Reworked per Codex's catch that GitHub issue assignees are users,
  not teams — so the earlier "use a team handle" comment was wrong.
  If CATALOG_TRACKER_ASSIGNEES is unset we now skip assignment
  entirely and emit a ::warning title=Drift notifications inactive::.
  No personal fallback; a deactivated handle would silently mute
  notifications. The warning surfaces in the workflow log every run
  until the secret is set.

#4 (LOW / clarification) — Deleted-sig case:
  Added an inline comment in the drift loop noting that an rsync
  that deletes skill.oms.sig alongside content edits registers the
  deletion in `git diff` (sig_changed matches → drift NOT flagged
  here), and the now-sigless skill falls through to the compliance
  step under "missing artifacts: skill.oms.sig". Defense in depth.

#1 (HIGH per Sayali) — README version inconsistency on revert:
  NOT addressed in this commit — verified against
  .github/scripts/regenerate-readme.sh on main: the README table
  is now "Product | Description | Skills" (PR #215, commit 1104de0
  on 2026-06-01 dropped the Source + Version columns). VERSIONS_FILE
  is declared but never consumed, and /tmp/sync-versions.txt is
  vestigial. So the exact "README will lie" risk Sayali raised is
  stale. Plan to send her the context separately and open a small
  follow-up PR to drop the dead sync-versions write + VERSIONS_FILE
  declaration so this trap doesn't catch a future reader.

Signed-off-by: Moshe Abramovitch <moshea@nvidia.com>
mosheabr added a commit that referenced this pull request Jun 8, 2026
#2 (MED) — gh issue create can skip tracker on bad assignee:
  The "Track dropped skills" step has continue-on-error: true, so a
  bad assignee wouldn't kill the whole sync — but it would skip
  tracker creation entirely, defeating the feature. Both branches
  (edit existing / create new) now follow the same tolerant pattern:
  create-without-assignees first, then loop --add-assignee in a
  trim + per-handle warning. A typo in CATALOG_TRACKER_ASSIGNEES no
  longer blocks the tracker from existing.

#3 (LOW) — Hardcoded personal fallback:
  Reworked per Codex's catch that GitHub issue assignees are users,
  not teams — so the earlier "use a team handle" comment was wrong.
  If CATALOG_TRACKER_ASSIGNEES is unset we now skip assignment
  entirely and emit a ::warning title=Drift notifications inactive::.
  No personal fallback; a deactivated handle would silently mute
  notifications. The warning surfaces in the workflow log every run
  until the secret is set.

#4 (LOW / clarification) — Deleted-sig case:
  Added an inline comment in the drift loop noting that an rsync
  that deletes skill.oms.sig alongside content edits registers the
  deletion in `git diff` (sig_changed matches → drift NOT flagged
  here), and the now-sigless skill falls through to the compliance
  step under "missing artifacts: skill.oms.sig". Defense in depth.

#1 (HIGH per Sayali) — README version inconsistency on revert:
  NOT addressed in this commit — verified against
  .github/scripts/regenerate-readme.sh on main: the README table
  is now "Product | Description | Skills" (PR #215, commit 1104de0
  on 2026-06-01 dropped the Source + Version columns). VERSIONS_FILE
  is declared but never consumed, and /tmp/sync-versions.txt is
  vestigial. So the exact "README will lie" risk Sayali raised is
  stale. Plan to send her the context separately and open a small
  follow-up PR to drop the dead sync-versions write + VERSIONS_FILE
  declaration so this trap doesn't catch a future reader.

Signed-off-by: Moshe Abramovitch <moshea@nvidia.com>
melo-gonzo pushed a commit to melo-gonzo/NVIDIAskills that referenced this pull request Aug 5, 2026
NVIDIA#2 (MED) — gh issue create can skip tracker on bad assignee:
  The "Track dropped skills" step has continue-on-error: true, so a
  bad assignee wouldn't kill the whole sync — but it would skip
  tracker creation entirely, defeating the feature. Both branches
  (edit existing / create new) now follow the same tolerant pattern:
  create-without-assignees first, then loop --add-assignee in a
  trim + per-handle warning. A typo in CATALOG_TRACKER_ASSIGNEES no
  longer blocks the tracker from existing.

NVIDIA#3 (LOW) — Hardcoded personal fallback:
  Reworked per Codex's catch that GitHub issue assignees are users,
  not teams — so the earlier "use a team handle" comment was wrong.
  If CATALOG_TRACKER_ASSIGNEES is unset we now skip assignment
  entirely and emit a ::warning title=Drift notifications inactive::.
  No personal fallback; a deactivated handle would silently mute
  notifications. The warning surfaces in the workflow log every run
  until the secret is set.

NVIDIA#4 (LOW / clarification) — Deleted-sig case:
  Added an inline comment in the drift loop noting that an rsync
  that deletes skill.oms.sig alongside content edits registers the
  deletion in `git diff` (sig_changed matches → drift NOT flagged
  here), and the now-sigless skill falls through to the compliance
  step under "missing artifacts: skill.oms.sig". Defense in depth.

NVIDIA#1 (HIGH per Sayali) — README version inconsistency on revert:
  NOT addressed in this commit — verified against
  .github/scripts/regenerate-readme.sh on main: the README table
  is now "Product | Description | Skills" (PR NVIDIA#215, commit 1104de0
  on 2026-06-01 dropped the Source + Version columns). VERSIONS_FILE
  is declared but never consumed, and /tmp/sync-versions.txt is
  vestigial. So the exact "README will lie" risk Sayali raised is
  stale. Plan to send her the context separately and open a small
  follow-up PR to drop the dead sync-versions write + VERSIONS_FILE
  declaration so this trap doesn't catch a future reader.

Signed-off-by: Moshe Abramovitch <moshea@nvidia.com>
Signed-off-by: Carmelo Gonzales <43048528+melo-gonzo@users.noreply.github.com>
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.

2 participants