feat: config wizard + 4 reference architectures + CI + outreach drafts + 4 new skills + i18n by OnlyTerp · Pull Request #8 · OnlyTerp/hermes-optimization-guide

OnlyTerp · 2026-04-17T05:03:23Z

Summary

Phase 4: highest-leverage follow-ups after PR #7. Moves the guide from comprehensive to canonical by adding an interactive config wizard, four detailed reference architectures, CI that actually enforces repo quality, launch-ready outreach drafts, four more installable skills, and entry-level localized READMEs in Chinese and Japanese.

Everything is still zero-dep / no-frameworks: the wizard is a single static HTML file, CI uses off-the-shelf actions plus one small Python validator, localized READMEs are just markdown.

Type

Docs / content update
New skill (skills/)
New config template (templates/config/) — only fixed quoting
Benchmark addition
Ecosystem entry
Infra template (.github/workflows/ci.yml, validator script)
Fix / typo / link

What landed

Interactive config wizard — `docs/wizard/`

Static HTML page, no frameworks. 8 questions (persona, default model, memory backend, gateways, MCP servers, approval layer, observability, cron tasks) → ready-to-drop config.yaml. copyYAML() and downloadYAML() helpers. Runs entirely in the browser; nothing uploaded anywhere. GitHub Pages friendly once enabled.

4 reference architectures — `docs/reference-architectures/`

Every blueprint has: who it's for, full parts list, install commands, full config.yaml, expected cost/month, scaling ceilings, graduation triggers.

homelab.md — privacy-first, on your own hardware, Tailscale mesh, Ollama, ~$0 infra
solo-developer.md — VPS + Telegram bot, cost-routed defaults, ~$25–70/mo
small-agency.md — 2–6 devs, per-client isolation via templated hermes@<name>.service units, compliance-friendly
road-warrior.md — phone-drives-cloud, Modal/Daytona/Fly on-demand, tar-pipe sync from Part 21

CI — `.github/workflows/ci.yml`

markdown-link-check on changed *.md (with .github/markdown-link-check.json allowlisting localhost / dynamic subdomains / example.com)
yamllint with a relaxed config (.github/yamllint.yml) that only catches real breakage, not formatting nits
validate_skills.py — every SKILL.md must have a valid YAML frontmatter with name, description (≥10 chars), non-empty when_to_use, and toolsets from a known whitelist
prettier --check advisory pass (non-blocking) to hint at drift

Ran locally before pushing: all 13 skills pass, yamllint clean.

Outreach drafts — `docs/outreach/`

Copy for Rob to fire off after this merges. Zero posted without approval.

launch-tweet-thread.md — 8-tweet thread + reply-suggestions, receipts-forward, no hype
hacker-news-post.md — Show HN title + first self-comment
reddit-localllama.md — homelab-angled post
nous-upstream-pr-body.md — Rob should open this, not me. Proposes a small "Community Guides" (plural) section in the Nous README. Designed to be low-review-burden and leave the door open for other guides to follow.
blog-post-long.md — ~1500-word dev.to / Substack format

4 new skills (total 13)

ops/daily-inbox-triage — morning sweep across email + Slack + Telegram DMs with classify + draft-reply + approval routing. Gemini Flash by default.
ops/hermes-weekly — Friday digest of NousResearch/hermes-agent merged PRs, classified (feature / fix / breaking / chore), ready to pipe to Telegram.
security/spam-trap — first-line untrusted-input filter for public channels. Cerebras Llama. Classifies GENUINE / SPAM / INJECTION / AMBIGUOUS and routes ambiguous to a quarantine profile.
dev/meeting-prep — 15-min-before-meeting 1-page brief. Pulls calendar + last threads + memory.

i18n — `README-zh.md`, `README-ja.md`

Entry-level summaries (the 21 parts themselves stay English). Large Hermes user bases in both communities per the v0.9 release notes. Links from the top of the main README.

Template YAML quoting fix

templates/config/{production,cost-optimized,security-hardened,telegram-bot}.yaml had unquoted ${VAR} substitutions inside flow mappings. yamllint flagged them; now quoted so every template is valid YAML as-is.

README

Skills badge 9 → 13
New CI badge
Language links row
Repo-map rows for docs/wizard/, docs/reference-architectures/, docs/outreach/

CHANGELOG + ROADMAP

Updated. The roadmap items "Interactive config wizard", "Reference architectures", "Integration tests", "Translations", "Hermes Weekly" moved from Queued → Done.

Checklist

Cross-links are relative (./partN-foo.md) and resolve
No secrets in any example — ${VAR} placeholders only
Dates / prices / PR numbers are current (marked 2026-04-17 where relevant)
For skills: security notes included; trust: posture documented (new skills explicitly set trust: untrusted where applicable and route risky actions through approval)
For templates: every non-obvious field is commented
CHANGELOG.md updated

After this merges

Things I am not doing without explicit approval:

Opening the upstream PR to NousResearch/hermes-agent (that draft is yours to send)
Posting to Twitter / HN / Reddit (all drafts are yours to post)

Things still queued on the ROADMAP: Astro Starlight docs site with full-text search, asciinema cast, Langfuse dashboard JSON export, dashboard-screenshots pass on parts 12/17/20.

Link to Devin session: https://app.devin.ai/sessions/42780dee7d0d4798b1910200a1f7280d
Requested by: @OnlyTerp

… skills + i18n - docs/wizard/ — interactive static config-builder (HTML+JS, GitHub Pages friendly) - docs/reference-architectures/ — 4 full blueprints (homelab, solo-dev, small-agency, road-warrior) - docs/outreach/ — launch drafts (tweet, HN, Reddit, upstream-Nous PR, long blog) - .github/workflows/ci.yml + validate_skills.py — markdown-link-check, yamllint, skill frontmatter linter, prettier advisory - skills/ — +4 (daily-inbox-triage, hermes-weekly, spam-trap, meeting-prep) — total 13 - README-zh.md, README-ja.md — localized entry summaries - templates/config/*.yaml — quoted ${VAR} inside flow mappings (valid YAML) - README: skills 9→13, language links, wizard/ref-arch/outreach rows, CI badge - CHANGELOG + ROADMAP updated Co-Authored-By: Rob <onerobby@gmail.com>

devin-ai-integration · 2026-04-17T05:03:26Z

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

Disable automatic comment and CI monitoring

Co-Authored-By: Rob <onerobby@gmail.com>

…YAML blocks Addresses Devin Review comment: the persona radio (Question 1) was only emitted as a YAML comment. Now it actually configures the starting state. - onchange="applyPersona(value)" on each persona radio - PERSONA_PRESETS mirrors templates/config/<persona>.yaml shape: minimum / telegram-bot / production / cost-optimized / security-hardened - applyPersona() sets default_model, memory, gateways, mcps, approval, obs, crons — then re-generates YAML automatically - generate() now emits persona-specific blocks: - security-hardened: security.mcp.{default_trust,require_allowlist,allow_sampling} - security-hardened + production: webhook require_signature + max_body_bytes, redaction.{memory_write,log}, quarantine profile - cost-optimized: routing.rules (5-rule cost ladder) Co-Authored-By: Rob <onerobby@gmail.com>

Devin Review flagged four mismatches between the PERSONA_PRESETS and the actual reference templates. Fixed: - minimum: memory none -> vector (templates/config/minimum.yaml:25 uses vector) - telegram-bot: memory vector -> lightrag (telegram-bot.yaml:43 uses lightrag) - cost-optimized: memory vector -> lightrag (cost-optimized.yaml:66 uses lightrag) - cost-optimized: obs helicone -> none (cost-optimized.yaml has no exporters; it's telemetry + alerts only) Now each preset actually mirrors its template, which is what the code comment at line 185-187 claims. Co-Authored-By: Rob <onerobby@gmail.com>

…plate) templates/config/cost-optimized.yaml uses llama-3.1-70b in all three places (classification, triage model, compress_model). The wizard was emitting llama-3.3-70b. Align. Co-Authored-By: Rob <onerobby@gmail.com>

All 5 outreach drafts and both i18n READMEs referenced the old count of 9 skills. This PR added 4 new skills (daily-inbox-triage, hermes-weekly, spam-trap, meeting-prep) bringing the total to 13. Updated files: - README-zh.md, README-ja.md (count + skill name list) - docs/outreach/blog-post-long.md - docs/outreach/hacker-news-post.md (count + enumerated list) - docs/outreach/launch-tweet-thread.md - docs/outreach/nous-upstream-pr-body.md - docs/outreach/reddit-localllama.md (count + enumerated list) Co-Authored-By: Rob <onerobby@gmail.com>

Devin Review caught a red issue: quoting ${TELEGRAM_OWNER_ID} inside a flow sequence (e.g. ["${TELEGRAM_OWNER_ID}"]) makes the YAML valid but changes the parsed type from int to str after env-var substitution. Telegram user IDs are numeric and Hermes compares them with strict type matching, so a string-typed list would reject all messages. Switched all 4 templates + 2 reference architectures + the wizard generator to block-sequence style: allowed_user_ids: - ${TELEGRAM_OWNER_ID} Verified with PyYAML that post-substitution type is int across all 4 templates. Also: localized READMEs (zh + ja) now use 'sudo bash' to match the main README, since vps-bootstrap.sh requires root. Co-Authored-By: Rob <onerobby@gmail.com>

Devin Review (red): the wizard previously emitted only the provider derived from the default model. But cost-optimized routing references cerebras, moonshot, and anthropic too — a wizard-generated config would fail at runtime when routing picked one of those. Refactored to a data-driven `EXTRA_PROVIDERS` map so each persona declares which providers its routing rules need. The generator then emits the union of (default-model provider, extra persona providers). - cost-optimized: anthropic, google, moonshot, cerebras - production: anthropic, google - security-hardened: anthropic (for the quarantine fallback) Also added cerebras to the providerKeys map (CEREBRAS_API_KEY). Co-Authored-By: Rob <onerobby@gmail.com>

Two Devin Review findings addressed: 1. skills/README.md catalog listed only 9 skills; PR added 4 new ones (daily-inbox-triage, hermes-weekly, spam-trap, meeting-prep). Appended rows for all 4 so the catalog matches the 13-skill count now quoted in README / CHANGELOG / outreach docs. 2. Wizard generator: when memory=lightrag is selected, the generated config references google/gemini-2.5-flash (LLM) and openai/text-embedding-3-small (embedding), but the provider-emission set only accounted for the default-model provider + routing-rule providers. For 3 of 5 personas this produced broken configs with missing embedding provider credentials. Added a lightrag branch that injects google + openai into providersToEmit. Co-Authored-By: Rob <onerobby@gmail.com>

devin-ai-integration bot assigned OnlyTerp Apr 17, 2026

fix: correct relative path from daily-inbox-triage to spam-trap skill

33703e0

Co-Authored-By: Rob <onerobby@gmail.com>

This comment was marked as resolved.

Sign in to view

OnlyTerp merged commit ca75386 into main Apr 17, 2026
5 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: config wizard + 4 reference architectures + CI + outreach drafts + 4 new skills + i18n#8

feat: config wizard + 4 reference architectures + CI + outreach drafts + 4 new skills + i18n#8
OnlyTerp merged 9 commits intomainfrom
devin/1776401488-phase4

OnlyTerp commented Apr 17, 2026 •

edited by devin-ai-integration bot

Loading

Uh oh!

devin-ai-integration bot commented Apr 17, 2026

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

OnlyTerp commented Apr 17, 2026 • edited by devin-ai-integration bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Type

What landed

Interactive config wizard — docs/wizard/

4 reference architectures — docs/reference-architectures/

CI — .github/workflows/ci.yml

Outreach drafts — docs/outreach/

4 new skills (total 13)

i18n — README-zh.md, README-ja.md

Template YAML quoting fix

README

CHANGELOG + ROADMAP

Checklist

After this merges

Uh oh!

devin-ai-integration bot commented Apr 17, 2026

🤖 Devin AI Engineer

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

OnlyTerp commented Apr 17, 2026 •

edited by devin-ai-integration bot

Loading

Interactive config wizard — `docs/wizard/`

4 reference architectures — `docs/reference-architectures/`

CI — `.github/workflows/ci.yml`

Outreach drafts — `docs/outreach/`

i18n — `README-zh.md`, `README-ja.md`