sawz/last30days-skill
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: vs mode N full passes + --competitors auto-discovery + (/Last30Days) title (#312)

Browse Source 
* feat: vs mode runs N full passes; --competitors wraps vs with auto-discovery

Unifies vs-mode and --competitors onto one fanout architecture. A topic
containing "vs" / "versus" now runs N full pipeline.run() calls in parallel
(reverting the one-pass latency optimization that removed per-entity
depth); --competitors becomes a SKILL.md-level shortcut where the hosting
reasoning model (Claude Code, Codex, Hermes, Gemini) discovers N peers via
its own WebSearch, runs Step 0.55 per entity, and invokes the engine with
a vs-topic + --competitors-plan JSON.

Changed:
- vs-mode: N full passes in parallel via fanout (was 1 merged pass).
- --competitors: SKILL.md shortcut for vs-mode-with-discovery. Engine flag
  kept for headless/cron use. LAW 7-style stderr reframed to lead with the
  hosting-model path (use WebSearch + --competitors-plan) instead of
  BRAVE_API_KEY. Footer BRAVE/SERPER nudge suppressed when --plan or
  --competitors-plan present (hosting model already has WebSearch).

Added:
- --competitors-plan JSON flag: per-entity {x_handle, x_related, subreddits,
  github_user, github_repos, context}. Accepts inline JSON or file path.
  subrun_kwargs_for helper is the single source of truth for per-entity
  kwargs — no closure-default fallthrough from main scope.
- Per-entity save files: each entity's sub-run produces its own
  {slug}-raw.md with a single-row Resolved Entities block.
- --polymarket-keywords filter for ambiguous single-token topics.

Fixed:
- test_competitor_subrun_isolation regression suite locks in 3.0.12's
  no-leak invariant (main flags do not inherit into peer sub-runs).
- Updates test_regression.py for the new comparison-mode payload shape.

Bumps plugin.json to 3.0.13. 1,219 tests passing.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: comparison title attribution — (Last 30 Days) → (/Last30Days)

User feedback on 3.0.13 dogfood runs (Kanye vs Drake, Mercer Island,
Figma): the comparison-mode synthesis title should attribute to the
slash command rather than restate the date range.

Three SKILL.md occurrences updated. Pure documentation change. Bumps to
3.0.14.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Matt Van Horn
2026-04-22 21:31:00 -07:00
committed by GitHub
parent 00d01933e0
commit 949bcf8942
18 changed files with 2022 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files
SKILL.md
+41 -23
View File
@@ -110,7 +110,7 @@ Replace `{VERSION}` with the installed plugin version (`jq -r '.version' "$SKILL
**Placement by query type:**
- GENERAL / NEWS / PROMPTING / RECOMMENDATIONS: badge on line 1, blank line 2, `What I learned:` on line 3, then bold-lead-in paragraphs
- COMPARISON: badge on line 1, blank line 2, `# {TOPIC_A} vs {TOPIC_B} [vs {TOPIC_C}]: What the Community Says (Last 30 Days)` on line 3, then Quick Verdict section
- COMPARISON: badge on line 1, blank line 2, `# {TOPIC_A} vs {TOPIC_B} [vs {TOPIC_C}]: What the Community Says (/Last30Days)` on line 3, then Quick Verdict section
---
@@ -128,7 +128,7 @@ These LAWs dominate every other rule in this file. If you find yourself about to
**LAW 2 - NO INVENTED TITLE LINE (with COMPARISON exception).** For QUERY_TYPE GENERAL, NEWS, PROMPTING, RECOMMENDATIONS: the first line of your synthesis body (after the badge and one blank line) is the prose label `What I learned:` on its own line. Not `What I learned about {Topic}`, not `{Topic} - Last 30 Days`, not `{Topic}: What People Are Saying`, not `# {Topic}`, not `The headline`, not `Why he is everywhere this month`. Nothing above `What I learned:` except the badge. If you are tempted to write a title or a `##`-prefixed section name, the rule is: the badge IS the title, and section headers are forbidden (see LAW 4).
**COMPARISON exception:** For QUERY_TYPE=COMPARISON (topics containing `vs` or `versus`), the title `# {TOPIC_A} vs {TOPIC_B} [vs {TOPIC_C}]: What the Community Says (Last 30 Days)` is REQUIRED, not a violation. Comparison queries do NOT use the `What I learned:` prose label at all.
**COMPARISON exception:** For QUERY_TYPE=COMPARISON (topics containing `vs` or `versus`), the title `# {TOPIC_A} vs {TOPIC_B} [vs {TOPIC_C}]: What the Community Says (/Last30Days)` is REQUIRED, not a violation. Comparison queries do NOT use the `What I learned:` prose label at all.
**Global-preference override:** The skill-authored template for GENERAL / NEWS / PROMPTING / RECOMMENDATIONS queries uses `**bold**` for KEY PATTERNS items and for mid-paragraph lead-ins. Do NOT strip this bold on the grounds of a personal "no bold" memory. The skill's voice contract is the formatting authority here.
@@ -570,43 +570,61 @@ Generated: {date} | Sources: Reddit, X, Bluesky, YouTube, TikTok, HN, Polymarket
## If QUERY_TYPE = COMPARISON
When the user asks "X vs Y", run ONE research pass with a comparison-optimized plan that covers both entities AND their rivalry. This replaces the old 3-pass approach (which took 13+ minutes and produced tangential content).
When the user asks "X vs Y" (or "X vs Y vs Z"), the engine fans out N full `pipeline.run()` calls in parallel — one per entity — each with its own Step 0.55-grade targeting. This restored the old N-pass architecture (reverted the one-pass latency optimization that removed per-entity depth); parallel execution keeps wall clock ≈ a single pass.
**IMPORTANT: Include BOTH X handles (`--x-handle={TOPIC_A_HANDLE} --x-related={TOPIC_B_HANDLE},{COMPANY_HANDLES},{COMMENTATOR_HANDLES}`), `--subreddits={RESOLVED_SUBREDDITS}`, `--tiktok-hashtags={RESOLVED_HASHTAGS}`, `--tiktok-creators={RESOLVED_TIKTOK_CREATORS}`, and `--ig-creators={RESOLVED_IG_CREATORS}` from Step 0.55. Omit any flag where the value was not resolved (empty).**
**MANDATORY per-entity resolution.** For each entity, resolve the full Step 0.55 stack (X handle, subreddits, GitHub user/repos, news context). Then assemble a `--competitors-plan` JSON mapping each entity to its targeting, and invoke the engine ONCE with the vs-topic string.
**Single pass with entity-aware subqueries:**
**Output shape per run:**
- Main topic saves to `{main-slug}-raw.md`.
- Each peer saves to `{peer-slug}-raw.md`.
- Stdout shows a merged comparison with the `## Head-to-Head` scaffold + per-entity Resolved Entities block.
**Invocation:**
```bash
"${LAST30DAYS_PYTHON}" "${SKILL_ROOT}/scripts/last30days.py" "{TOPIC_A} vs {TOPIC_B}" --emit=compact --save-dir="${LAST30DAYS_MEMORY_DIR}" --save-suffix=v3 --plan 'COMPARISON_PLAN_JSON' --x-handle={TOPIC_A_HANDLE} --x-related={TOPIC_B_HANDLE},{COMPANY_A_HANDLE},{COMPANY_B_HANDLE},{COMMENTATOR_HANDLES} --subreddits={RESOLVED_SUBREDDITS} --tiktok-hashtags={RESOLVED_HASHTAGS} --tiktok-creators={RESOLVED_TIKTOK_CREATORS} --ig-creators={RESOLVED_IG_CREATORS}
"${LAST30DAYS_PYTHON}" "${SKILL_ROOT}/scripts/last30days.py" "{TOPIC_A} vs {TOPIC_B} vs {TOPIC_C}" \
--emit=compact \
--save-dir="${LAST30DAYS_MEMORY_DIR}" \
--save-suffix=v3 \
--x-handle={TOPIC_A_HANDLE} \
--subreddits={TOPIC_A_SUBS} \
--competitors-plan '{
"{TOPIC_B}": {"x_handle":"{TOPIC_B_HANDLE}","subreddits":["{TOPIC_B_SUB_1}","{TOPIC_B_SUB_2}"],"github_user":"{TOPIC_B_GH}","context":"{TOPIC_B_CONTEXT}"},
"{TOPIC_C}": {"x_handle":"{TOPIC_C_HANDLE}","subreddits":["{TOPIC_C_SUB_1}"],"github_user":"{TOPIC_C_GH}","context":"{TOPIC_C_CONTEXT}"}
}'
```
**The `--plan` JSON for comparisons should include 3-4 subqueries:**
1. **Head-to-head:** `"{TOPIC_A} vs {TOPIC_B}"` - catches rivalry content, direct comparisons
2. **Entity A news:** `"{TOPIC_A} news {MONTH} {YEAR}"` - catches entity-specific developments
3. **Entity B news:** `"{TOPIC_B} news {MONTH} {YEAR}"` - catches entity-specific developments
4. (Optional) **Domain context:** `"{COMPANY_A} {COMPANY_B} {DOMAIN} news"` - catches industry context (e.g., "OpenAI Anthropic AI news")
Topic A (the main topic, first in the vs-string) uses outer `--x-handle`, `--x-related`, `--subreddits`, `--github-user`, `--github-repo`, `--tiktok-*`, `--ig-creators` as usual. Topics B and C get their targeting from `--competitors-plan` entries (keyed by entity name, case-insensitive).
ALL subqueries include ALL sources. The fusion engine handles deduplication across subqueries. **At least one subquery MUST include YouTube-specific search terms** (e.g., "{PERSON} interview 2026", "{PRODUCT_A} vs {PRODUCT_B} review") to ensure YouTube content is found. Without YouTube-specific terms, the engine may only find 0-1 videos for comparison queries.
**Step 0.55 for N entities.** The same pre-research protocol that applies to a single-entity topic applies to EACH entity in a vs-run. For N=3, that means 3 WebSearches for X handles, 3 for subreddits, 3 for GitHub, 3 for news context — or equivalent batched queries. A `## Resolved Entities` block with dashes for any entity means you skipped Step 0.55 for that one. Re-run with a corrected plan.
Then do WebSearch for: `{TOPIC_A} vs {TOPIC_B} comparison {YEAR}` and `{TOPIC_A} vs {TOPIC_B} which is better` and `{COMPANY_A} vs {COMPANY_B} news {MONTH} {YEAR}`.
**Then do WebSearch supplements** for: `{TOPIC_A} vs {TOPIC_B} comparison {YEAR}` and `{TOPIC_A} vs {TOPIC_B} which is better` — these catch rivalry articles that per-entity passes might not surface.
**Skip the normal Step 1 below** - go directly to the comparison synthesis format (see "If QUERY_TYPE = COMPARISON" in the synthesis section).
**COMPARISON TABLE SCAFFOLD (engine-emitted, pass through verbatim):** For comparison topics, the engine's compact output includes a `## Head-to-Head Comparison` block with an empty markdown table (columns = entities, rows = axes like "Core pitch", "Who it's for", "Community stance", "Trajectory") plus a "Choose X if / Choose Y if" prose block. Your synthesis MUST include this block verbatim with filled cells, positioned between the narrative and the emoji-tree footer. Keep each cell to 5-15 words. Use ' - ' (hyphen with spaces) not em-dashes inside cells. The block is the canonical comparison output shape - do not invent your own table structure.
**COMPARISON TABLE SCAFFOLD (engine-emitted, pass through verbatim):** For comparison topics, the engine's compact output includes a `## Head-to-Head` block with an empty markdown table (columns = entities, rows = axes like "What it is", "Community sentiment", "Trajectory"). Your synthesis MUST include this block verbatim with filled cells, positioned between the narrative and the emoji-tree footer. Keep each cell to 5-15 words. Use ' - ' (hyphen with spaces) not em-dashes inside cells.
### Competitor mode (`--competitors`)
When the user passes `--competitors` on a single-entity topic, the engine auto-discovers 1-6 peer entities and fans out the full pipeline over the topic plus each competitor in parallel. Example: `last30days Kanye West --competitors` resolves to a 3-way comparison against Drake and Kendrick Lamar; `last30days OpenAI --competitors=3` resolves against Anthropic, xAI, and Google Gemini.
`--competitors` is a SKILL.md-level shortcut for vs-mode with auto-discovery. The engine flag itself just signals intent; YOU (the hosting reasoning model) do the discovery and Step 0.55 via your own WebSearch tool, then invoke the vs-topic path above.
**Flag surface:**
- `--competitors` (bare) - discover and compare against 2 peers (3-way comparison: original + 2).
- `--competitors=N` - discover N peers (range 1..6; out-of-range clamps with a stderr warning).
- `--competitors-list="A,B,C"` - skip discovery and use the explicit list. Implies `--competitors`.
**The four-step protocol:**
1. **Discover peers** via WebSearch: `"{topic} competitors"` / `"{topic} alternatives"`. Pick N=2 by default (match the flag's default), N=argument value if the user passed `--competitors=N`.
2. **Run Step 0.55 for the main topic AND each peer** — same protocol you use for a single-entity topic, just N times. X handle, subreddits, GitHub, news context, per entity.
3. **Build the vs-topic string**: `"{main} vs {peer1} vs {peer2}"`.
4. **Invoke the engine** with the vs-topic, `--competitors-plan` JSON covering both peers (and the main topic if you want to override the outer flags), and the outer `--x-handle`/`--subreddits`/`--github-*` for the main topic.
**Discovery path:** web search plus deterministic text mining (same backends as `--auto-resolve`). No internal LLM call. When no web search backend is configured and no list is passed, the engine emits a LAW 7-style stderr telling the hosting reasoning model to generate the list and re-invoke with `--competitors-list="A,B,C"`, then exits non-zero.
**Flag surface (engine):**
- `--competitors` (bare) - signals the hosting model to discover 2 peers (3-way total).
- `--competitors=N` - N peers (1..6; out-of-range clamps with stderr warning).
- `--competitors-list="A,B,C"` - minimum escape hatch; names only, no per-entity targeting. Peer sub-runs fall back to planner defaults (visibly thinner data).
- `--competitors-plan '{entity: {x_handle, subreddits, github_user, github_repos, context}}'` - full per-entity targeting; implies vs-mode; preferred.
- `--polymarket-keywords "kw1,kw2"` - disambiguate Polymarket for ambiguous single-token topics ("Warriors" → `nba,gsw,golden-state`).
**Sub-run behavior:** each entity runs `pipeline.run()` in parallel inheriting the main run's `--quick` / `--deep` / `--web-backend` / `--days`. Topic-specific overrides (`--x-handle`, `--subreddits`, `--github-user`, `--github-repo`) apply to the main topic only - competitor sub-runs use planner defaults. Per-entity failures degrade to a warning; the run continues as long as at least 2 entities survive.
**Why --competitors-plan over --competitors-list:** without per-entity handles/subs, peer sub-runs run with deterministic single-word planner queries and produce visibly thinner evidence than the main topic. The Resolved Entities block in stdout makes the gap visible — dashes for a peer = you skipped its Step 0.55.
**Output:** one comparison report covering all entities, reusing the same `## Head-to-Head` scaffold as explicit `A vs B` topics. Synthesis contract identical to the COMPARISON query type above.
**Engine-internal auto-resolve (headless fallback):** if the engine detects BRAVE_API_KEY / EXA_API_KEY / SERPER_API_KEY / PARALLEL_API_KEY / OPENROUTER_API_KEY, it runs its own per-entity `resolve.auto_resolve()` before each sub-run. The hosting-model path does NOT need those keys — you are the WebSearch. The engine's auto-resolve is the cron/CI fallback for when no reasoning model is driving.
**Output:** one `{slug}-raw.md` per entity in `--save-dir` plus the merged comparison on stdout. Synthesis contract identical to the vs-mode protocol above.
---
@@ -1187,7 +1205,7 @@ Voice contract LAWs 1, 3, 5 apply to comparisons unchanged (no `Sources:` block,
```
🌐 last30days v{VERSION} · synced {YYYY-MM-DD}
# {TOPIC_A} vs {TOPIC_B} [vs {TOPIC_C}]: What the Community Says (Last 30 Days)
# {TOPIC_A} vs {TOPIC_B} [vs {TOPIC_C}]: What the Community Says (/Last30Days)
## Quick Verdict