last30days-skill/docs/plans/2026-02-14-feat-merge-openclaw-variant-plan.md

title, type, date

title	type	date
feat: Merge OpenClaw variant into main repo	feat	2026-02-14

feat: Merge OpenClaw Variant into Main Repo

Overview

Consolidate the last30days-openclaw project into last30days-skill-private so there's one unified Python engine powering both the main skill (Claude Code / Codex) and an "open" variant with watchlist, briefing, history, and built-in web search. The open variant also gets YouTube and Bird CLI — features the main project already has but openclaw was built before they existed.

Problem Statement / Motivation

Right now there are two separate repos with diverging codebases:

• last30days-skill-private (main, Feb 14) — YouTube, vendored Bird, better scoring/normalization, Codex compat. But no built-in web search APIs and no persistence layer.
• last30days-openclaw (Feb 10) — SQLite store, watchlist, briefings, 3 web search backends (Parallel AI, Brave, OpenRouter). But frozen without YouTube or latest engine improvements.

They share ~80% of the same scripts/lib/ files but are drifting apart. Maintaining two codebases is unsustainable.

Goal: One repo, one Python engine, two SKILL.md variants. Install once, works everywhere.

Proposed Solution

Use last30days-skill-private as the base (it's 4 days newer with better code) and port the OpenClaw-exclusive features in:

What gets ported from OpenClaw

File	What it does	Destination
scripts/store.py	SQLite research accumulator (WAL, FTS5, dedup)	scripts/store.py
scripts/watchlist.py	Topic watchlist CLI (add/remove/list/run)	scripts/watchlist.py
scripts/briefing.py	Morning briefing generator (daily/weekly)	scripts/briefing.py
scripts/lib/brave_search.py	Brave Search API (free tier, 2K/mo)	scripts/lib/brave_search.py
scripts/lib/parallel_search.py	Parallel AI search (LLM-optimized)	scripts/lib/parallel_search.py
scripts/lib/openrouter_search.py	OpenRouter/Sonar Pro search	scripts/lib/openrouter_search.py
references/research.md	One-shot research instructions	variants/open/references/research.md
references/watchlist.md	Watchlist mode instructions	variants/open/references/watchlist.md
references/briefing.md	Briefing mode instructions	variants/open/references/briefing.md
references/history.md	History query instructions	variants/open/references/history.md

What gets upgraded in the ported code

• store.py: No changes needed — it's self-contained SQLite, works as-is
• watchlist.py: Remove OpenClaw cron-specific code, make cron setup generic (launchd on macOS, systemd on Linux, or manual cron)
• briefing.py: No changes needed
• scripts/lib/env.py: Merge OpenClaw's web search key support (PARALLEL_API_KEY, BRAVE_API_KEY, OPENROUTER_API_KEY) and has_web_search_keys() / get_web_search_source() functions into main's env.py. Drop the OpenClaw config loader (~/.openclaw/openclaw.json) — just use env vars and ~/.config/last30days/.env
• scripts/last30days.py: Add OpenClaw's _search_web() function so the script can do web search natively when API keys are available (instead of always delegating to the assistant)

What gets DROPPED from OpenClaw

File	Why
scripts/cron_setup.py	Too OpenClaw-platform-specific. Replace with generic scheduling docs.
OpenClaw config loader in env.py	~/.openclaw/openclaw.json path is platform-specific. Use env vars instead.
.clawhubignore	OpenClaw marketplace artifact, not needed in unified repo

New file: Open variant SKILL.md

Create variants/open/SKILL.md — the multi-mode skill with command routing:

variants/open/
├── SKILL.md              # Router: watch, briefing, history, or one-shot
├── references/
│   ├── research.md       # One-shot research instructions
│   ├── watchlist.md      # Watchlist management instructions
│   ├── briefing.md       # Briefing mode instructions
│   └── history.md        # History query instructions
└── context.md            # Agent memory (user preferences, source quality)

The open variant's SKILL.md points to {baseDir}/scripts/last30days.py (same engine) but adds the router and reference file system. It also adds the --store flag for persistence.

How YouTube and Bird CLI get added to the open variant

They're already in scripts/lib/youtube_yt.py and scripts/lib/vendor/bird/. The open variant's SKILL.md just needs to mention YouTube in its description and the research.md reference file gets the YouTube stats line in the output format. No code changes needed — the Python engine already supports all four sources.

Technical Considerations

File Structure After Merge

last30days-skill-private/
├── SKILL.md                          # Main skill (Claude Code / Codex)
├── agents/openai.yaml                # Codex discovery (existing)
├── variants/
│   └── open/
│       ├── SKILL.md                  # Open variant with routing
│       ├── references/
│       │   ├── research.md
│       │   ├── watchlist.md
│       │   ├── briefing.md
│       │   └── history.md
│       └── context.md
├── scripts/
│   ├── last30days.py                 # Unified engine (+ native web search)
│   ├── store.py                      # SQLite accumulator (from openclaw)
│   ├── watchlist.py                  # Watchlist CLI (from openclaw, genericized)
│   ├── briefing.py                   # Briefing generator (from openclaw)
│   └── lib/
│       ├── ... (existing files)
│       ├── brave_search.py           # NEW from openclaw
│       ├── parallel_search.py        # NEW from openclaw
│       ├── openrouter_search.py      # NEW from openclaw
│       ├── youtube_yt.py             # Existing
│       └── vendor/bird/              # Existing
└── README.md                         # Updated with open variant docs

Installation for open variant users

# Claude Code (main skill — unchanged)
git clone https://github.com/mvanhorn/last30days-skill.git ~/.claude/skills/last30days

# Open variant (with watchlist, briefings, history)
git clone https://github.com/mvanhorn/last30days-skill.git ~/.claude/skills/last30days
# Then in Claude Code settings, point skill to variants/open/SKILL.md
# OR symlink:
ln -sf ~/.claude/skills/last30days/variants/open/SKILL.md ~/.claude/skills/last30days-open/SKILL.md

env.py merge strategy

Main's env.py is the base. Add from OpenClaw:

• Three new key names: PARALLEL_API_KEY, BRAVE_API_KEY, OPENROUTER_API_KEY
• has_web_search_keys() function
• get_web_search_source() function — returns 'parallel', 'brave', or 'openrouter'
• get_available_sources() update to include web-search-capable modes

last30days.py merge strategy

Main's last30days.py is the base. Add from OpenClaw:

• _search_web() function that calls the appropriate web search backend
• --store CLI flag to persist findings to SQLite
• --diagnose CLI flag for source availability diagnostics
• Web results integration into the existing report pipeline (normalize → score → dedupe → render)

Keep main's:

• YouTube integration
• Phase 2 supplemental search
• 3-tier Reddit fallback
• Better error handling
• Minimum result guarantee

Portable path resolution (already done)

The main SKILL.md already has portable path resolution (from Codex compat work):

for dir in "." "${CLAUDE_PLUGIN_ROOT:-}" "$HOME/.claude/skills/last30days" ...

The open variant's SKILL.md uses {baseDir} which resolves to the skill root. Both approaches work — we just need to make sure the open variant's references use {baseDir} consistently.

Acceptance Criteria

• scripts/store.py ported and working (SQLite creates on first use)
• scripts/watchlist.py ported with generic scheduling (no OpenClaw cron dependency)
• scripts/briefing.py ported and generates daily/weekly briefings
• scripts/lib/brave_search.py ported and functional
• scripts/lib/parallel_search.py ported and functional
• scripts/lib/openrouter_search.py ported and functional
• scripts/lib/env.py updated with web search key support
• scripts/last30days.py has native _search_web() + --store + --diagnose
• variants/open/SKILL.md exists with command routing (watch, briefing, history, research)
• variants/open/references/*.md — all 4 reference files ported
• Open variant mentions YouTube in description and research output format
• Open variant uses same portable path resolution as main
• Main SKILL.md behavior is unchanged (zero regressions)
• python3 scripts/last30days.py "test topic" --mock --emit=compact still works
• python3 scripts/last30days.py "test topic" --diagnose shows source availability
• README documents open variant installation and usage

Dependencies & Risks

Risk	Likelihood	Mitigation
OpenClaw's store.py has import dependencies we don't have	Low	store.py uses only stdlib (sqlite3, json, datetime). Self-contained.
Web search backends need API keys to test	Medium	Each has a --mock or dry-run path. Test with real keys if available, mock otherwise.
watchlist.py depends on OpenClaw cron API	High	Known — strip cron_setup.py dependency, replace with generic docs for launchd/systemd/crontab.
Open variant SKILL.md is too long (>500 lines)	Medium	Use reference file pattern (already planned). Router SKILL.md stays under 100 lines.
env.py merge introduces regressions	Low	Main's env.py is well-tested. Additive changes only — new keys, new functions.
Two SKILL.md files = maintenance burden	Low	They serve different purposes. Main is simple one-shot. Open adds routing. Core engine is shared.

Files to Create/Modify

New Files

• variants/open/SKILL.md — Open variant router (~100 lines)
• variants/open/references/research.md — One-shot research instructions (from openclaw, updated with YouTube)
• variants/open/references/watchlist.md — Watchlist management instructions (from openclaw)
• variants/open/references/briefing.md — Briefing mode instructions (from openclaw)
• variants/open/references/history.md — History query instructions (from openclaw)
• variants/open/context.md — Agent memory template
• scripts/store.py — SQLite accumulator (from openclaw, as-is)
• scripts/watchlist.py — Watchlist CLI (from openclaw, genericized)
• scripts/briefing.py — Briefing generator (from openclaw, as-is)
• scripts/lib/brave_search.py — Brave Search API (from openclaw)
• scripts/lib/parallel_search.py — Parallel AI search (from openclaw)
• scripts/lib/openrouter_search.py — OpenRouter/Sonar Pro search (from openclaw)

Modified Files

• scripts/lib/env.py — Add web search key support (~30 lines added)
• scripts/last30days.py — Add _search_web(), --store, --diagnose (~80 lines added)
• README.md — Add open variant section (~20 lines)

Total scope: ~12 new files (mostly copied), ~130 lines of new code in existing files.

References

Internal

• OpenClaw plan: /Users/mvanhorn/last30days-openclaw/docs/plans/2026-02-10-feat-openclaw-last30days-skill-plan.md (989 lines, comprehensive spec)
• Codex compat plan: docs/plans/2026-02-14-feat-codex-skill-compatibility-plan.md (portable paths, platform-neutral text)
• OpenClaw source: /Users/mvanhorn/last30days-openclaw/

Key files to port

• store.py: /Users/mvanhorn/last30days-openclaw/scripts/store.py (20KB, SQLite with FTS5)
• watchlist.py: /Users/mvanhorn/last30days-openclaw/scripts/watchlist.py (10KB)
• briefing.py: /Users/mvanhorn/last30days-openclaw/scripts/briefing.py (8KB)
• brave_search.py: /Users/mvanhorn/last30days-openclaw/scripts/lib/brave_search.py (6KB)
• parallel_search.py: /Users/mvanhorn/last30days-openclaw/scripts/lib/parallel_search.py (4KB)
• openrouter_search.py: /Users/mvanhorn/last30days-openclaw/scripts/lib/openrouter_search.py (7KB)
• env.py (openclaw version): /Users/mvanhorn/last30days-openclaw/scripts/lib/env.py (9KB — has web search key functions)