agentleFS
jyjejyje/hermes-agent-helm/.claude/skills/review-upstream/SKILL.md

review-upstream

Skill18 starsChanged 31 days agoMITView on GitHub

Review Hermes Agent stable releases against this Helm chart using original release notes, tagged source, and small per-release checks. Use for upstream reviews, image-bump compatibility reviews, and authorized upstream-review issue filing.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use a clean snapshot when unrelated edits are present.

    line 14
  2. Distinctive:

    Read deployment, upgrade, persistence, security and configuration changes.

    line 22
  3. Very specific:

    Read the filtering contract in advise-image-bump.py, especially SYSTEM_PROMPT.

    line 38
  4. Distinctive:

    Read the relevant implementation before running its mode:

    line 53
  5. Distinctive:

    Check semantic duplicates first: the script only checks exact open titles.

    line 88
  6. Very specific:

    Use source: manual for an agent's original-note review, not NIM attribution.

    line 95

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Review upstream

Review upstream changes that require chart action. Review-only requests do not authorize issue creation, image bumps, workflow dispatch, PRs, or merges.

Original evidence and per-release checks

  1. Record the chart commit, appVersion, requested release window and dirty work. Use a clean snapshot when unrelated edits are present. If no window is given, use current appVersion through the requested target and state the baseline.
  2. Inventory formal stable GitHub releases separately from Docker tags. Paginate discovery and follow AGENTS.md's tag policy. Retrieve the original body of every release using gh api repos/NousResearch/hermes-agent/releases/tags/<tag>. Retain bodies, URLs, dates and draft/prerelease flags. Missing originals mean unreviewed, not no findings; never silently substitute summaries.
  3. Read deployment, upgrade, persistence, security and configuration changes. Account for cumulative notes, reverted changes and later correction notices. State omitted sections. Inspect relevant source at the exact tag, not moving main. For environment-table audits, use hermes-agent-env-rule.
  4. Perform small checks for each tag: image availability/architectures, Helm lint and rendering with an explicit image override, and relevant chart-used paths, migration/auth interfaces or runtime pins. Exercise both controller types when relevant. Choose checks proportional to the changes, not a fixed large cluster matrix. Record pass/fail/not-run per release.
  5. Separate static checks from container startup, persisted-PVC migration and authenticated live checks. A render does not prove the image exists; an image does not prove a safe release. Do not start credentialed flows without scope. This review window is not a historical image-support commitment.

Chart relevance and existing work

Read the filtering contract in advise-image-bump.py, especially SYSTEM_PROMPT. Freeform config/env passthrough already exposes new keys. Flag broken existing examples, changed defaults/persistence contracts, invalid documented assumptions, security impact or actual resource changes, not bare new keys or desktop-only features.

Compare current templates, values, examples, English/Korean docs, open/closed issues and active PRs. Each finding needs priority, scope, original source links, acceptance checks and evidence limits. Reference covered work instead of filing duplicates. Keep permanent CI follow-ups separate under AGENTS.md. Do not infer a chart major bump solely from the upstream version label.

Shared scripts and authorized filing

Read the relevant implementation before running its mode:

  • advise-image-bump.py is the optional NIM advisor. Inspect its output against original notes and report omitted input. Manual review needs no NIM key. Skipped/failed output is not a clean review.
  • file-image-bump-issues.py writes issues with shared titles, priority labels, triage guidance and an exact-open-title duplicate guard. Run only when issue filing is authorized.
  • review-upstream-releases.py inventories formal stable releases, retains originals and small per-tag checks, and records only completed reviews by original-note digest. Inspect the committed manual baseline and cached ledger; neither means findings are fixed. --plan-only provides a read-only inventory preview without model or issue calls.
  • cron-fetch-image.yaml runs formal reviews independently of image bump gates. An authorized dispatch with review_only=true cannot propose an image bump; review_from selects an inclusive baseline and force_review=true rechecks recorded digests. It can still create issues and consume NIM requests. Cache eviction causes backfill, and missing notes or omitted critical sections require manual review.

For an authorized manual filing, prepare and inspect this JSON contract:

{
  "source": "manual",
  "items": [{
    "priority": "medium",
    "title": "Concrete chart follow-up",
    "detail": "Evidence, affected files, requested change, acceptance checks and source links.",
    "upstream_ref": "Release tag and section or upstream PR"
  }]
}

Check semantic duplicates first: the script only checks exact open titles. Then run from the repository root:

python3 .github/scripts/release/file-image-bump-issues.py <review.json> <target-tag> jyje/hermes-agent-helm

Use source: manual for an agent's original-note review, not NIM attribution. Group independent review windows by target tag. After partial failure inspect created issues before retrying. Do not close issues or implement fixes unless requested. Never copy the shared script into the skill directory.

Finish with per-release results, findings or verified issue URLs, and explicit unverified areas. Agent/CI-only tooling needs no Changeset; user-facing chart fixes follow the repository Changeset and locale rules.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.