Master decision thread: every open question for agentlab.in v1

[harshitsinghbhandari]

This is the master decision-tracking thread for agentlab.in. The vision is set (community publishing platform for AI agent infrastructure knowledge — patterns, playbooks, deep dives). This thread captures every open question that needs a human decision before we build.

Reply inline (quote each question) with your decision and I'll lock it in. Anything you don't answer, I'll pick a sensible default and call it out in a follow-up.

---

1. Auth & Identity

1. GitHub OAuth only — confirm no email/password fallback?

   • Default: yes, GitHub-only. No email fallback.

2. Username scheme on agentlab.in/@username

   • (a) Always mirror the GitHub login exactly (no custom handle)
   • (b) Default to GitHub login but let users pick a custom handle once
   • (c) Fully custom handles, independent of GitHub login

3. Profile data source

   • (a) Always pull live from GitHub (bio, avatar, location) — no override
   • (b) Pull from GitHub at signup, then user can edit anything
   • (c) Hybrid: avatar always from GitHub, bio editable

4. Minimum GitHub account age to post? (anti-spam guard)

   • (a) No minimum
   • (b) 7 days
   • (c) 30 days
   • (d) Account must have ≥1 public repo

5. Org accounts — can a GitHub org (e.g. @anthropic) be an author, or strictly individual humans only?

---

2. Content Types & Structure

6. Pattern posts — enforce structured sections?

   • (a) Free-form markdown, template shown as a suggestion only
   • (b) Soft structure: prefilled headings (Problem / Structure / Trade-offs / Related), user can delete them
   • (c) Hard structure: dedicated fields rendered into a fixed layout (like a recipe card)

7. Playbook posts — same question. Free-form, soft template, or hard structure?

8. Deep Dive posts — same question. I assume fully free-form, confirm?

9. Can a post change type after publishing? (e.g. Deep Dive → Pattern)

   • (a) Yes, anytime
   • (b) Yes, within 24h of publish
   • (c) No, type is permanent

10. Required fields for any post

    • Title — required, obviously
    • Body — required
    • Type — required
    • Tags — required (min 1)? optional?
    • Summary/dek (1-line subtitle) — required? optional?
    • Cover image — required? optional? none?

---

3. Writing Experience

11. Editor

    • (a) Plain markdown textarea + live preview (simplest, ship in a day)
    • (b) Rich Tiptap/Lexical editor that outputs markdown
    • (c) Split-pane markdown editor (CodeMirror left, preview right)

12. MDX support? (lets authors embed React components — Callouts, Tweets, etc.)

    • (a) No, plain markdown only
    • (b) Yes, but only an allowlist of components we ship
    • (c) Full MDX (risky — XSS surface)

13. Drafts

    • (a) Yes, auto-saved drafts before publish
    • (b) No drafts — write it, ship it
    • (c) Drafts only via "Save as draft" button (no auto-save)

14. Edit history / versioning

    • (a) No history — current version only
    • (b) Track edits, show "edited" timestamp, no version browser
    • (c) Full version history visible to readers (like Wikipedia)

15. Code blocks — pick a syntax highlighter

    • (a) Shiki (VS Code-grade, slower build)
    • (b) Prism (lighter, faster)
    • (c) Highlight.js (simplest)

16. Math / LaTeX support (KaTeX)? yes / no

17. Mermaid diagrams? yes / no

18. Image uploads

    • (a) Supabase Storage (in-house, free tier)
    • (b) Cloudinary (better transforms, free tier)
    • (c) No uploads — paste image URLs only

19. Slug generation

    • (a) Auto from title, user cannot edit
    • (b) Auto-suggest from title, user can override
    • (c) User must enter slug manually

20. URL structure

    • (a) /patterns/trust-gate, /playbooks/legacy-portals, /deep-dives/ollama-json
    • (b) /@harshit/trust-gate (author-first, type implied in metadata)
    • (c) /p/trust-gate (single namespace, type shown on page)

---

4. Tags & Taxonomy

21. Tag model

    • (a) Free-form tags, anyone can create any tag
    • (b) Curated tag list, we approve new tags
    • (c) Free-form but with auto-merge of obvious duplicates (auth = authentication)

22. Max tags per post? (3, 5, unlimited?)

23. Tag landing pages at /tags/security? yes / no

24. Featured / canonical tags shown on homepage? (e.g. security, local-first, orchestration, memory, evals) — give me the list you want pinned.

---

5. Discoverability

25. Homepage layout — pick one

    • (a) Chronological feed (newest first)
    • (b) Curated/editor's picks above, chronological below
    • (c) By type: three columns (latest patterns / playbooks / deep dives)
    • (d) Mixed feed ranked by a heat score (recency + engagement)

26. Search backend

    • (a) Postgres full-text (free, good enough for <10k posts)
    • (b) Meilisearch self-hosted (better UX, needs infra)
    • (c) Algolia (best UX, paid)

27. RSS feed? yes / no (yes is the obvious answer, confirming)

28. Email digest / newsletter — ship at launch or later?

29. Sitemap.xml + robots.txt — yes (assumed). Confirm we want full indexing.

---

6. Engagement & Social

30. Comments on posts

    • (a) No comments — pure publishing platform
    • (b) Native flat comments
    • (c) Native threaded comments
    • (d) Embed GitHub Discussions per post (zero infra, GitHub-native)

31. Reactions (👍 💡 🔥 etc.) — yes / no

32. Bookmarks / save for later — yes / no

33. Follow authors — yes / no

34. View counts — show publicly, show to author only, or don't track?

35. Cross-references between patterns

    • (a) Manual [[Pattern Name]] wiki-style autolinks
    • (b) Manual hyperlinks only
    • (c) Both manual links + auto-generated "Referenced by" backlinks on the target

36. "Related posts" at bottom of each post

    • (a) Tag-overlap based (automatic)
    • (b) Manual — author picks 3 related posts
    • (c) None

---

7. Moderation & Trust

37. Moderation model

    • (a) You're the only mod
    • (b) You + 1-2 trusted co-mods
    • (c) Community flagging + you adjudicate

38. Report button on posts/comments — yes / no

39. Content policy — do you want one written? (spam, hate, plagiarism rules)

40. Author can delete their own posts — yes (assumed), or no (posts are canonical and stay)?

41. Plagiarism / attribution policy

    • When someone "distills Karpathy's tweets into a pattern" — what's the rule?
    • (a) Must credit original source with link, post authored under distiller's name
    • (b) Must credit + must contact original author for OK
    • (c) Distilled posts are clearly badged ("Distilled from @karpathy")

---

8. Bootstrap / Seed Strategy

42. Authorship of seed posts

    • (a) All under your personal account (@harshitsinghbhandari)
    • (b) Split: your own work under you, distilled work under a dedicated @agentlab curator account
    • (c) Create placeholder profiles for cited authors (risky — looks like impersonation)

43. The 50-60 seed list — do you want me to draft the full list now, or seed in batches as we build?

44. Permission for distilling public-but-attributed content (Karpathy tweets, Anthropic guides)

    • (a) Just attribute heavily, assume fair use
    • (b) Reach out for explicit OK before publishing
    • (c) Only distill from things with explicit reuse licenses (CC, blog posts inviting summaries)

---

9. Visual Design & Brand

45. Dark theme reference — pick the closest vibe

    • (a) Vercel.com (pure black, sharp)
    • (b) Linear.app (near-black, soft)
    • (c) Stripe Press (warm dark, editorial)
    • (d) Anthropic.com (cream/serif, NOT dark — confirm if you want light theme instead)

46. Typography direction

    • (a) Serif body (editorial — like NYT, Stripe Press)
    • (b) Sans body, mono accents (technical — like Linear)
    • (c) Mono everywhere (hacker — like Berkshire Hathaway-meets-Vercel)

47. Light mode — ship both dark + light, or dark only at launch?

48. Logo / wordmark — do you have one, want me to draft options, or skip and use wordmark-only for now?

49. Code block theme — once syntax highlighter is picked, which palette? (One Dark, Dracula, GitHub Dark, Tokyo Night, custom?)

---

10. Profile Pages

50. What's on /@username? check all that apply

    • Avatar + bio from GitHub
    • Posts list (all / by type)
    • Linked GitHub repos (auto-fetched, top N by stars)
    • Pinned posts (user-selected)
    • Stats (post count, total views, follower count)
    • Social links (Twitter, personal site)
    • "Currently working on" status line

51. Username change — allow it? what happens to old URLs (redirect, 404)?

---

11. Technical / Infra

52. Supabase region — pick one (closest to you/audience)

    • (a) ap-south-1 (Mumbai) — closest to you
    • (b) us-east-1 (N. Virginia) — closest to US audience
    • (c) eu-west-2 (London) — neutral global

53. Image storage — Supabase Storage (default) or external?

54. Analytics

    • (a) Plausible (privacy-friendly, paid)
    • (b) PostHog (free tier, more features)
    • (c) Vercel Analytics (built-in, basic)
    • (d) None

55. Error tracking — Sentry (free tier) or skip at launch?

56. CI/CD — Vercel previews on every PR (assumed). Confirm.

57. Branching model — feature branches → PR → main → auto-deploy to production? Or staging branch in between?

---

12. Launch Strategy

58. Launch mode

    • (a) Public from day 1 — anyone can sign up immediately
    • (b) Soft launch: site is live and readable, sign-ups gated by invite for first 2 weeks
    • (c) Stealth: site lives at a non-indexed URL until seed content is done, then flip the switch

59. Launch channels — HN, ProductHunt, your Twitter, Composio Slack/Discord, Anthropic-adjacent communities? Pick which we'll target on launch day.

60. Pre-launch waitlist / coming-soon page? yes / no

---

13. Project Governance

61. Is agentlab.in itself open source?

    • (a) Yes, MIT — anyone can contribute to the platform code
    • (b) Yes, source-available but not open contribution
    • (c) Closed source

62. Accept platform code contributions from the community? yes / no

63. License on user-submitted content

    • (a) CC BY 4.0 (anyone can reuse with attribution)
    • (b) CC BY-SA 4.0 (reuse with attribution + share-alike)
    • (c) Author retains all rights, agentlab gets perpetual hosting license

---

14. Monetization (future, but decide direction now)

64. Monetization path
    • (a) Never — community-owned forever, funded by you/donations
    • (b) Sponsorships (clearly-labeled "Sponsored by X" posts or banners)
    • (c) Paid tier later (premium features, no paywalled posts)
    • (d) Decide later — no monetization decisions until 1k DAU

---

15. Anything I Missed?

65. Anything from the vision doc that you want to change or refine before we build?

66. Anything not in the vision doc that you want included?

67. Hard deadlines? (any event you want to launch at — IIT Bombay event, a conference, hackathon demo?)

---

How to reply: quote each line you care about and write your answer. For anything you skip, I'll default to the safest/simplest option and call out what I picked.

[harshitsinghbhandari]

1. Yes Github Only. Possibly will never add any other options.
2. Username cannot be changed, will be github username only. agentlab.in/{id} will mean github user github.com/abcd will be always agentlab.in/abcd
3. you can change anything except display-name and username
4. At least 30 days and at least 1 public repo
5. yes org accounts will and must be allowed.
6. Soft Structure for posts is great idea.
7. Playbook will be hard structure, we'll discuss
8. Deep Dives, will have 1-2 mandatory sections like topic and summary something
9. No Type is permanent. You can copy to each other but will not convert.
10. I genuinely dont know so whatever you feel is best.
11. split pane (updating 30 sec server side)
12. Yes, with our allowed components
13. Drafts only via "Save as draft" button (no auto-save)
14. Track edits, show "edited" timestamp, no version browser
15. Prism (lighter, faster)
16. not yet
17. yes
18. I would prefer urls only and but yeah add images also, size limit 2MB
19. automatic slug from title, user cannot change
20. we will use X format -> https://x.com/RomanSaini/status/2059971904438665560 ->
    https://agentlab.in/harshitsinghbhandari/{posts/dive/playbook}/{slug}
21. Curated list, approval for new tags and nested tags also tag/sub-tag system
22. 5
23. Yes tag landing pages
24. We'll discuss later
25. By type: three columns (For You page / playbooks / deep dives), Mixed feed ranked by a heat score (recency + engagement), we can X's algorithm for you page
26. Postgres full-text, because we are not going 10K posts for at least 3-4 months.
27. yes obvio
28. Not yet
29. Yes full
30. Native threaded comments
31. Only Like (simple as instagram)
32. Yes, this will be our main great thing
33. Yes, obvio
34. Show to author
35. I don't get the question
36. Tag-overlap based (automatic)
37. Community flagging + you I can block and delete posts too.
38. Yes obviously
39. Yes obviously
40. Yes
41. Must credit original source with link, post authored under distiller's name, content is their issue, they post they are responsibly agent lab isn't responsible, strong content policy, I will draft myself and we'll imrpovise.
42. Split: your own work under you, distilled work under a dedicated @agentlab curator account
43. we will start slowly, will start with only 1-2 posts and slowly move.
44. Distill only, content policy Core policy
    Allowed -> summaries explanations critiques comparisons architectural breakdowns “what I learned from” independent reimplementations short quotations with attribution
    Not allowed -> full reposts excessive copying paywalled content dumps entire threads mirrored raw scraped archives proprietary leaks
45. Vercel.com (pure black, sharp)
46. Mono everywhere (hacker — like Berkshire Hathaway-meets-Vercel)
47. dark + light
48. skip and use wordmark-only for now
49. GitHub Dark
50. Avatar + bio from GitHub
    Posts list (all / by type)
    Linked GitHub repos (auto-fetched, top N by stars)
    Pinned posts (user-selected)
    Stats (post count, total views, follower count)
51. No username change (anyone will fucking name to karpathy and we can't have that, so we need github actual)
52. ap-south-1 (Mumbai)
53. Supabase Storage (default)
54. Vercel Analytics (built-in, basic)
55. skip at launch
56. WOrks, vercel preview
57. staging branch in between
58. immediate signups obvio
59. I'll do this myself, we'll think later, it's gonna take at least 2 weeks to build properly.
60. No
61. closed source at starting, we'll open source later not now
62. No, not at starting
63. CC BY 4.0 (anyone can reuse with attribution)
64. Decide later — no monetization decisions until 1k DAU
65. No
66. We will make it very AI Agent friendly, great api endpoints and documentations, but not right now
67. No

[harshitsinghbhandari]

Locked everything else in. A few items still need a decision before I can start building — numbered so you can reply inline.

---

A. Playbook hard structure (your Q7 follow-up)

What are the mandatory sections for a Playbook? My proposal:

1. Environment / Target — what this playbook teaches an agent to operate in
2. Prerequisites — what the agent must already have/know
3. Core Instructions — the actual AGENTS.md-style body (this is where most of the content lives)
4. Safety / Failure Modes — what to refuse, what to escalate
5. Examples — at least one worked example

Yes / change / add / remove?

B. Deep Dive mandatory sections (your Q8 follow-up)

You said "topic and summary something." My proposal: exactly 2 required sections at the top:

1. TL;DR (3-5 lines, the takeaway)
2. The Question (what this deep dive is actually exploring)

Rest is free-form. Sound good?

C. URL type segments (your Q20)

You wrote {posts/dive/playbook}. Confirming the three literal URL segments. My read:

• Patterns → /posts/ (or do you want /patterns/?)
• Deep Dives → /dive/
• Playbooks → /playbook/

Example: agentlab.in/harshitsinghbhandari/posts/trust-gate

Confirm the three exact words. Plural vs singular? posts vs patterns?

D. Required fields on every post (your Q10)

You said "whatever you feel is best." My proposal:

• Required: Title, Body, Type, Summary (1-line dek), at least 1 Tag
• Optional: Cover image (URL or upload, ≤2MB per your Q18)

Yes / change?

E. Cross-references (your Q35 — restating)

Patterns often reference each other ("Trust Gate composes with Memory Taxonomy"). Two questions:

1. Should authors be able to write [[Pattern Name]] and have it auto-link to that pattern's page? (wiki-style)
2. Should each pattern page automatically show a "Referenced by" section listing other posts that link to it? (backlinks)

(a) Both yes, (b) just [[autolinks]], (c) just backlinks, (d) neither — let authors do plain markdown links manually.

F. Editor save semantics (Q11 + Q13 reconciliation)

You picked split-pane with "updating 30 sec server side" AND "Save as draft button, no auto-save." Confirming intent:

• The 30s server-side update = compile preview / sync state to server, but does not persist as a draft
• Draft persistence only happens when user clicks Save as draft

Right? Or did you mean the 30s updates DO persist as a draft (i.e. effectively auto-save every 30s)?

G. "For You" algorithm (Q25)

You referenced X's algorithm. For v1, do you want:

• (a) A real personalization model (engagement signals, embeddings, etc.) — complex, weeks of work
• (b) A heat-ranked feed labeled "For You" (recency + likes + bookmarks + tag affinity from your own posts) — ship in days, can evolve later
• (c) Just chronological under the "For You" label and iterate later

H. Featured / pinned tags (Q24 — we said later, flagging it's still TBD)

Not blocking, but flag for whenever you want to pick. Suggested starter set: security, local-first, orchestration, memory, evals, tooling, prompting, multi-agent. Adopt / edit / wait?

I. Content policy text (Q39 / Q41)

You said you'll draft. Tracking that it must ship before the public sign-up button goes live. No action needed from me until you share the draft.

---

Once A–G are locked, I'll draft the database schema, route map, and a build order for v1 and post it as the next thread.

[harshitsinghbhandari]

A. Not needed to decide right now, can be changed later, your proposed is fine; Environment / Target — what this playbook teaches an agent to operate in
Prerequisites — what the agent must already have/know
Core Instructions — the actual AGENTS.md-style body (this is where most of the content lives)
Safety / Failure Modes — what to refuse, what to escalate
-> works for me. Drop examples
B. Sounds Good
C. post, dive and playbook seems perfect.
D. Yes, works
E. Both yes
F. We will save the drafts in local memory of the browser, every 30sec, browser cache or local memory cleared clears the drafts also. Once we reach a lot of reach, then we add a paid option to sync drafts to server too. But that's a later thing.
G. (b)
H. Your starter-set works
I. leave it for now, I will draft, I will also draft terms of use, privacy policy (we're using github data) and a support url