sawz/last30days-skill
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
feat/scrape-source-opt-in
last30days-skill/docs/plans/2026-03-02-chore-open-pr-triage-plan.md
T
Matt Van Horn 32992834ee Merge PR #48: feat: add Xiaohongshu source + Reddit public fallback 
- Xiaohongshu search via local MCP service (opt-in, zero impact if service not running)
- Reddit public JSON fallback (works with zero API keys)
- Reddit priority: ScrapeCreators -> OpenAI -> public fallback
- Updated env.py: Reddit always available via public fallback

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-07 16:11:35 -08:00

8.1 KiB
Raw Permalink Blame History

title, type, status, date
title type status date
chore: open PR triage - build or close decision for all 5 open PRs chore active 2026-03-02

Open PR Triage

Decision record for each of the 5 open PRs on mvanhorn/last30days-skill. For each: build/merge, cherry-pick, or close with explanation.

PR #26 - Add HN, YouTube, and Product Hunt sources + --search flag

Author: wkbaran | Size: +3402/-49 | Date: 2026-02-15

What it adds

  • Hacker News source (Algolia API, no auth)
  • YouTube source (YouTube Data API v3, optional key)
  • Product Hunt source (GraphQL API, optional key)
  • --search=SOURCES flag (e.g. --search=reddit,hn,yt)
  • 58 new tests

What's already on main

  • HN: ✅ already merged (scripts/lib/hackernews.py)
  • YouTube: ✅ already merged (scripts/lib/youtube_yt.py)
  • Product Hunt: ❌ NOT on main
  • --search flag: ❌ NOT on main

Decision: CHERRY-PICK (--search flag only, skip Product Hunt)

The HN and YouTube modules in this PR are earlier versions than what's on main. We should NOT take those. Product Hunt is not a priority source. The one missing piece worth landing:

--search=SOURCES flag - lets callers (including agent mode) specify which sources to run. Useful for --search=hn,polymarket or --search=reddit,x focus runs.

Plan

  • Cherry-pick just tests/test_search_flag.py and the --search argument wiring in scripts/last30days.py
  • Skip scripts/lib/producthunt.py and all PH-related code entirely
  • Do NOT take their hackernews.py, youtube.py, render.py, normalize.py, schema.py - our versions are newer
  • Update SKILL.md to document --search flag
  • Comment on PR thanking wkbaran - his HN/YouTube work was an inspiration for the sources we ended up building. Acknowledge that specifically.

Acceptance Criteria

  • python3 scripts/last30days.py "AI coding tools" --search=hn,reddit runs only HN + Reddit
  • python3 scripts/last30days.py "test" --search=x,web skips Reddit
  • test_search_flag.py passes (adapt to our test patterns, no PH references)
  • SKILL.md updated with --search flag docs
  • Sync via bash scripts/sync.sh

PR #24 - Adding Codex compatibility

Author: el-analista | Size: +228/-29 | Date: 2026-02-11

What it adds

  • agents/openai.yaml Codex discovery metadata
  • Portable script path resolution in SKILL.md (multi-install-path loop)
  • Platform-neutral UI text (assistant instead of Claude-specific wording)
  • LAST30DAYS_CACHE_DIR env override in scripts/lib/cache.py
  • LAST30DAYS_OUTPUT_DIR env override in scripts/lib/render.py
  • X/Bird query noise stripping + last-chance retry in scripts/lib/bird_x.py
  • New tests: test_bird_x.py, test_cache.py, test_render.py

What's already on main

  • Codex auth: ✅ already merged (via PR #38)
  • Multi-path script resolution: ✅ already in SKILL.md (the for dir in loop)
  • Platform-neutral wording: ❌ NOT systematically applied
  • Cache/output dir overrides: ❌ NOT on main
  • Bird X improvements: ✅ partially (we have other bird_x fixes but may be missing noise stripping + retry)

Decision: REVIEW AND PARTIAL CHERRY-PICK

The env var overrides (LAST30DAYS_CACHE_DIR, LAST30DAYS_OUTPUT_DIR) are genuinely useful for sandboxed/containerized Codex environments. The bird_x noise stripping and last-chance retry may improve X search quality and are low-risk additions.

The platform-neutral wording change ("assistant" instead of Claude references) should be skipped — SKILL.md is Claude-specific by design.

Plan

  • Read the diff carefully against our current cache.py, render.py, bird_x.py
  • Cherry-pick the LAST30DAYS_CACHE_DIR and LAST30DAYS_OUTPUT_DIR env overrides
  • Cherry-pick the bird_x noise stripping + retry logic if it doesn't conflict with our existing bird_x changes
  • Skip agents/openai.yaml — we have our own multi-path resolution
  • Skip platform-neutral wording changes
  • Comment on PR thanking contributor and explaining what landed

Acceptance Criteria

  • LAST30DAYS_CACHE_DIR=/tmp/test python3 scripts/last30days.py "test" --mock writes cache to /tmp/test
  • LAST30DAYS_OUTPUT_DIR=/tmp/out python3 scripts/last30days.py "test" --mock writes output to /tmp/out
  • Existing tests pass after cherry-pick
  • test_cache.py and test_render.py adapted and passing

PR #14 - Simplify to WebSearch-first, make API keys optional

Author: thangman1 | Size: +97/-168 | Date: 2026-02-01

What it does

Removes the Reddit/X Python search engine as the primary data source. Repositions Claude Code's built-in WebSearch as the "default" mode. Strips engagement metrics (upvotes, likes, repost counts) from output. Removes mode detection logic (Full Mode / Partial Mode / Web-Only Mode).

Decision: CLOSE - do not merge

This PR inverts the core value proposition of last30days. The skill's differentiation is real engagement data from Reddit threads and X posts — upvotes, likes, reposts — that WebSearch cannot provide. Stripping that out produces a worse tool than just asking Claude to search the web, which anyone can already do.

The author's intent (lower barrier to entry, no API key required) is valid, but the right solution is making HN + Polymarket work without any API key (they already do), and making OPENAI_API_KEY easier to obtain — not removing the Reddit/X engine.

Action

  • Comment on PR explaining why we're closing it
  • Acknowledge the valid friction point (API key setup) and point to HN + Polymarket as the zero-config sources
  • Close PR

PR #10 - OpenRouter API integration

Author: thetechreviewer | Size: +1029/-16 | Date: 2026-01-28

What it adds

OpenRouter as an alternative to OpenAI for the Reddit discovery search. Allows using any model available on OpenRouter instead of just OpenAI models.

What's already on main

  • scripts/lib/openrouter_search.py: ✅ ALREADY ON MAIN
  • Wired into scripts/last30days.py: ✅ ALREADY ON MAIN (the backend == "openrouter" path)

Decision: CLOSE - already merged

The OpenRouter integration that this PR introduced is already on main. It arrived via internal work that post-dated this PR. The PR is stale.

Action

  • Comment on PR: "Thanks for this — OpenRouter support is already on main (landed via internal work). Closing as incorporated."
  • Close PR

PR #5 - Add support for Codex auth with OpenAI Responses API

Author: jblwilliams | Size: +358/-66 | Date: 2026-01-27

What it adds

  • JWT-based Codex auth with chatgpt_account_id
  • Codex endpoint routing (https://chatgpt.com/backend-api/codex/responses)
  • SSE handling for streaming Codex responses
  • Typed auth status/source dataclass
  • Codex fallback model chain

What's already on main

All of this is already on main. The Codex auth system landed via PR #37 (iliaal:codex-auth-merged), which in turn came in with PR #38. This PR (#5) predates that work and covers the same ground.

Decision: CLOSE - already incorporated

Action

  • Comment on PR: "Thanks for this early work on Codex auth! The same feature landed on main via PR #37 (from a separate contributor who built on similar ideas). The JWT decode, Codex endpoint routing, SSE parsing, and typed auth dataclass are all live. Closing as incorporated."
  • Close PR

Summary Table

PR Author Decision Reason
#26 wkbaran Cherry-pick --search flag not on main; HN/YouTube already there; skip Product Hunt
#24 el-analista Cherry-pick Cache dir overrides + bird_x improvements worth landing; Codex auth already there
#14 thangman1 Close Removes engagement data, inverts core value proposition
#10 thetechreviewer Close OpenRouter already on main
#5 jblwilliams Close Codex auth already on main via PR #37/38

Implementation Order (if proceeding)

  1. Close PR #5, #10, #14 with comments (no code changes needed)
  2. Cherry-pick PR #24 pieces (bird_x + cache/render env overrides)
  3. Cherry-pick PR #26 pieces (Product Hunt + --search flag)
  4. Sync and test
  5. Push to upstream
Reference in New Issue View Git Blame Copy Permalink