AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary
Python ai-promptsai-skillblueskyclaudeclaude-codeclawhubdeep-researchhackernewsinstagramopenclawpolymarketrecencyredditresearchsocial-mediatiktoktrendstwitterweb-searchyoutube
chore(release): bump version to 3.22.0 (#1085)
github-actions[bot]fcebe321c22e
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.jsonindex e7da0489..ce0cbd1f 100644--- a/.claude-plugin/marketplace.json+++ b/.claude-plugin/marketplace.json@@ -11,7 +11,7 @@{"name": "last30days","description": "Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and 5+ more sources. AI agent scores by upvotes, likes, and real money - not editors.",- "version": "3.21.1",+ "version": "3.22.0","author": {"name": "Matt Van Horn","url": "https://github.com/mvanhorn"diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.jsonindex 6844609a..9e2a2fd5 100644--- a/.claude-plugin/plugin.json+++ b/.claude-plugin/plugin.json@@ -1,6 +1,6 @@{"name": "last30days",- "version": "3.21.1",+ "version": "3.22.0","description": "Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and 5+ more sources. AI agent scores by upvotes, likes, and real money - not editors.","author": {"name": "Matt Van Horn",diff --git a/.codex-plugin/plugin.json b/.codex-plugin/plugin.jsonindex 53e277b1..038c0943 100644--- a/.codex-plugin/plugin.json+++ b/.codex-plugin/plugin.json@@ -1,6 +1,6 @@{"name": "last30days",- "version": "3.21.1",+ "version": "3.22.0","description": "Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and the web.","author": {"name": "Matt Van Horn",diff --git a/.grok-plugin/marketplace.json b/.grok-plugin/marketplace.jsonindex a063c8f1..785a3189 100644--- a/.grok-plugin/marketplace.json+++ b/.grok-plugin/marketplace.json@@ -9,7 +9,7 @@{"name": "last30days","description": "Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and 5+ more sources. AI agent scores by upvotes, likes, and real money - not editors.",- "version": "3.21.1",+ "version": "3.22.0","category": "productivity","source": {"source": "url",diff --git a/.grok-plugin/plugin.json b/.grok-plugin/plugin.jsonindex 62da1af9..37aa53cb 100644--- a/.grok-plugin/plugin.json+++ b/.grok-plugin/plugin.json@@ -1,6 +1,6 @@{"name": "last30days",- "version": "3.21.1",+ "version": "3.22.0","description": "Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and 5+ more sources. AI agent scores by upvotes, likes, and real money - not editors.","author": {"name": "Matt Van Horn",diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 39492861..051ec595 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -9,6 +9,34 @@ This project uses [towncrier](https://towncrier.readthedocs.io/). Upcoming notes<!-- towncrier release notes start -->+## [3.22.0] - 2026-08-31++### Added++- Telegram public channel source: opt-in via `--telegram-sources=handle1,handle2` or `TELEGRAM_SOURCES` + `INCLUDE_SOURCES=telegram`. Named public channels only (no keyword discovery); fetches recent posts via ScrapeCreators API and scores by views, reactions, and topic relevance. ([#990](https://github.com/mvanhorn/last30days-skill/issues/990))+- Add an explicit `--web-backend=parallel-mcp` option for anonymous hosted Parallel web search.+- Reddit threads with the most upvotes and comments now keep their place: each Reddit stream holds slots for its top three on-topic threads by engagement before per-stream truncation, and the fused candidate pool reserves slots (quick 2, default 3, deep 4) for the highest-engagement entity-grounded Reddit candidates. A 16K-upvote thread with weak title overlap was previously cut behind one-upvote posts. Topics whose primary entity starts with a generic word ("ai", "new") require the higher 0.25 relevance floor for these slots.++### Changed++- Keyless Reddit comment enrichment now covers 4 / 8 / 12 threads per subquery at quick / default / deep depth (was 3 / 5 / 8) and keeps up to 12 top comments per thread (was 10), now that repeat fetches are memoized and paced.+- The emoji-tree footer no longer appends `⚠ partial after N items: HTTP 429 ... (run doctor for fixes)` outcome text to any source line; source lines carry counts and engagement only. `--emit=compact` stdout also drops the `Some sources failed` / `Some sources returned partial results` warning lines and the `## Source Errors` block. The model-facing `## Partial Coverage` note stays, and the saved raw file, `--emit=json`, and `doctor --postmortem` keep full `source_status`, `warnings`, and `errors_by_source`.++### Fixed++- Keyless Reddit comment-enrichment slots now go to the most-commented threads within each entity-priority tier, so a high-discussion thread no longer misses `## Top Community Comments` coverage while near-empty threads consume the scarce slots. ([#906](https://github.com/mvanhorn/last30days-skill/issues/906))+- Keyless Reddit now paces unauthenticated reddit.com requests at 1 req/sec (configurable via `LAST30DAYS_REDDIT_KEYLESS_RATE`) and retries a 429'd RSS or listing sub-request once, so routine runs no longer drop those lanes as `partial` HTTP 429. ([#985](https://github.com/mvanhorn/last30days-skill/issues/985))+- A source that delivered items but lost some sub-requests (a swallowed 429 or 403 on one lane) now records `ok` with a detail line such as `3 sub-requests rate-limited (HTTP 429)` instead of `partial`; `doctor --postmortem` shows that detail on the Succeeded line. Adapter-declared leg failures (for example a Perplexity `both` run whose agent leg failed) still brand the source partial.+- Fusion and report finalize now identify a thread by its normalized URL, not its per-stream item id. When the same Reddit thread arrives from two subquery streams (both labelled `R1`), the candidate keeps the copy that carries `top_comments`, `comment_insights`, and the verified counts instead of discarding it as a duplicate, and `items_by_source` holds one entry per thread. This is what silently emptied `## Top Community Comments` on multi-subquery runs.+- Keyless Reddit GETs are memoized for the life of one command, and concurrent requesters for the same URL share the in-flight fetch. Subreddit listing partials, listing RSS feeds, arctic supplements, and shreddit comment pages were fetched once per subquery (four times on a typical run) because the Reddit lane is dispatched with the raw topic every time; a four-subquery run now issues roughly 50 reddit.com requests instead of ~184, which is what kept tripping the anonymous rate limit.+- Keyless Reddit RSS and listing fetches now size their per-future timeouts from the shared bucket's queue depth instead of a fixed 20 seconds. At 1 req/s with four subquery streams sharing the bucket, the fixed timeout expired while a fetch was still waiting for its token and the feed was silently dropped.+- Keyless Reddit comment scraping now drops bot authors (RemindMeBot, AutoModerator, `WikiTextBot`-style camelCase names, and `-bot`/`_bot` suffixed accounts), so a "I will be messaging you in 3 days" reply can no longer occupy a Top Community Comments slot. Ordinary usernames ending in a lowercase "bot" (Talbot, abbot) are unaffected. Cherry-picked from [#1034](https://github.com/mvanhorn/last30days-skill/pull/1034) (rebase of [#907](https://github.com/mvanhorn/last30days-skill/pull/907)).+- Rate-limit retries are bounded: a `Retry-After` / `x-ratelimit-reset` wait is capped at 60 seconds and an epoch-style reset (GitHub) is converted to a delta, so a 429 can no longer park a worker or the main thread for minutes. The keyless Reddit memo elects one new owner when an in-flight fetch fails instead of letting every waiter refetch, its gate wait and the comment-enrichment budget account for the shared bucket's queue, the Reddit pool reservation survives a small pool crowded by many sources, stream keepers respect `--max-per-source`, a source whose swallowed sub-requests all 429ed is not retried against the same host and is reported as rate-limited rather than `no-results` when filtering leaves it empty.+- Reddit 429 backoff now honours `x-ratelimit-reset`, not just `Retry-After`. Reddit's anonymous search/RSS endpoints reply to a 429 with `x-ratelimit-reset: 42` and no `Retry-After`, so the retry fell through to exponential backoff (3s, 5s, 9s) — every step shorter than the window Reddit actually requires. Each retry re-429'd, the budget drained, and Reddit was reported as a dead source when it was only being asked too early. Both `http.fetch_url` and `reddit_public` now read either header through the new `http.retry_delay_from_headers` helper.+- Sources that return items are no longer branded `auth-failed` / `partial` when a swallowed lane-level HTTP failure (e.g. Reddit shreddit partials 403-ing on datacenter egress) is captured by the pipeline sink. The transport-failure outcome still surfaces when nothing was delivered, so `doctor` prescriptions are unaffected.+- `## Top Community Comments` and `## Best Takes` now draw from every cluster that clears the relevance floor, not only the eight clusters shown in `## Ranked Evidence Clusters`, so a top-voted comment on a lower-ranked thread still reaches the brief.++## [3.21.1] - 2026-08-18### Changeddiff --git a/changelog.d/+fusion-keeps-enriched-copy.fixed.md b/changelog.d/+fusion-keeps-enriched-copy.fixed.mddeleted file mode 100644index 2b3f3640..00000000--- a/changelog.d/+fusion-keeps-enriched-copy.fixed.md+++ /dev/null@@ -1 +0,0 @@-Fusion and report finalize now identify a thread by its normalized URL, not its per-stream item id. When the same Reddit thread arrives from two subquery streams (both labelled `R1`), the candidate keeps the copy that carries `top_comments`, `comment_insights`, and the verified counts instead of discarding it as a duplicate, and `items_by_source` holds one entry per thread. This is what silently emptied `## Top Community Comments` on multi-subquery runs.diff --git a/changelog.d/+parallel-search-mcp.added.md b/changelog.d/+parallel-search-mcp.added.mddeleted file mode 100644index fa03cb2a..00000000--- a/changelog.d/+parallel-search-mcp.added.md+++ /dev/null@@ -1 +0,0 @@-Add an explicit `--web-backend=parallel-mcp` option for anonymous hosted Parallel web search.diff --git a/changelog.d/+quiet-footer.changed.md b/changelog.d/+quiet-footer.changed.mddeleted file mode 100644index d3eede90..00000000--- a/changelog.d/+quiet-footer.changed.md+++ /dev/null@@ -1 +0,0 @@-The emoji-tree footer no longer appends `⚠ partial after N items: HTTP 429 ... (run doctor for fixes)` outcome text to any source line; source lines carry counts and engagement only. `--emit=compact` stdout also drops the `Some sources failed` / `Some sources returned partial results` warning lines and the `## Source Errors` block. The model-facing `## Partial Coverage` note stays, and the saved raw file, `--emit=json`, and `doctor --postmortem` keep full `source_status`, `warnings`, and `errors_by_source`.diff --git a/changelog.d/+reddit-bot-comments.fixed.md b/changelog.d/+reddit-bot-comments.fixed.mddeleted file mode 100644index 572011c2..00000000--- a/changelog.d/+reddit-bot-comments.fixed.md+++ /dev/null@@ -1 +0,0 @@-Keyless Reddit comment scraping now drops bot authors (RemindMeBot, AutoModerator, `WikiTextBot`-style camelCase names, and `-bot`/`_bot` suffixed accounts), so a "I will be messaging you in 3 days" reply can no longer occupy a Top Community Comments slot. Ordinary usernames ending in a lowercase "bot" (Talbot, abbot) are unaffected. Cherry-picked from [#1034](https://github.com/mvanhorn/last30days-skill/pull/1034) (rebase of [#907](https://github.com/mvanhorn/last30days-skill/pull/907)).diff --git a/changelog.d/+reddit-engagement-floor.added.md b/changelog.d/+reddit-engagement-floor.added.mddeleted file mode 100644index baceec44..00000000--- a/changelog.d/+reddit-engagement-floor.added.md+++ /dev/null@@ -1 +0,0 @@-Reddit threads with the most upvotes and comments now keep their place: each Reddit stream holds slots for its top three on-topic threads by engagement before per-stream truncation, and the fused candidate pool reserves slots (quick 2, default 3, deep 4) for the highest-engagement entity-grounded Reddit candidates. A 16K-upvote thread with weak title overlap was previously cut behind one-upvote posts. Topics whose primary entity starts with a generic word ("ai", "new") require the higher 0.25 relevance floor for these slots.diff --git a/changelog.d/+reddit-enrich-limits.changed.md b/changelog.d/+reddit-enrich-limits.changed.mddeleted file mode 100644index 0f8aa3b1..00000000--- a/changelog.d/+reddit-enrich-limits.changed.md+++ /dev/null@@ -1 +0,0 @@-Keyless Reddit comment enrichment now covers 4 / 8 / 12 threads per subquery at quick / default / deep depth (was 3 / 5 / 8) and keeps up to 12 top comments per thread (was 10), now that repeat fetches are memoized and paced.diff --git a/changelog.d/+reddit-ok-with-lane-detail.fixed.md b/changelog.d/+reddit-ok-with-lane-detail.fixed.mddeleted file mode 100644index 06125a13..00000000--- a/changelog.d/+reddit-ok-with-lane-detail.fixed.md+++ /dev/null@@ -1 +0,0 @@-A source that delivered items but lost some sub-requests (a swallowed 429 or 403 on one lane) now records `ok` with a detail line such as `3 sub-requests rate-limited (HTTP 429)` instead of `partial`; `doctor --postmortem` shows that detail on the Succeeded line. Adapter-declared leg failures (for example a Perplexity `both` run whose agent leg failed) still brand the source partial.diff --git a/changelog.d/+reddit-queue-aware-timeouts.fixed.md b/changelog.d/+reddit-queue-aware-timeouts.fixed.mddeleted file mode 100644index 6ed618cb..00000000--- a/changelog.d/+reddit-queue-aware-timeouts.fixed.md+++ /dev/null@@ -1 +0,0 @@-Keyless Reddit RSS and listing fetches now size their per-future timeouts from the shared bucket's queue depth instead of a fixed 20 seconds. At 1 req/s with four subquery streams sharing the bucket, the fixed timeout expired while a fetch was still waiting for its token and the feed was silently dropped.diff --git a/changelog.d/+reddit-ratelimit-reset.fixed.md b/changelog.d/+reddit-ratelimit-reset.fixed.mddeleted file mode 100644index f5fdce5b..00000000--- a/changelog.d/+reddit-ratelimit-reset.fixed.md+++ /dev/null@@ -1 +0,0 @@-Reddit 429 backoff now honours `x-ratelimit-reset`, not just `Retry-After`. Reddit's anonymous search/RSS endpoints reply to a 429 with `x-ratelimit-reset: 42` and no `Retry-After`, so the retry fell through to exponential backoff (3s, 5s, 9s) — every step shorter than the window Reddit actually requires. Each retry re-429'd, the budget drained, and Reddit was reported as a dead source when it was only being asked too early. Both `http.fetch_url` and `reddit_public` now read either header through the new `http.retry_delay_from_headers` helper.diff --git a/changelog.d/+reddit-review-hardening.fixed.md b/changelog.d/+reddit-review-hardening.fixed.mddeleted file mode 100644index f650dc52..00000000--- a/changelog.d/+reddit-review-hardening.fixed.md+++ /dev/null@@ -1 +0,0 @@-Rate-limit retries are bounded: a `Retry-After` / `x-ratelimit-reset` wait is capped at 60 seconds and an epoch-style reset (GitHub) is converted to a delta, so a 429 can no longer park a worker or the main thread for minutes. The keyless Reddit memo elects one new owner when an in-flight fetch fails instead of letting every waiter refetch, its gate wait and the comment-enrichment budget account for the shared bucket's queue, the Reddit pool reservation survives a small pool crowded by many sources, stream keepers respect `--max-per-source`, a source whose swallowed sub-requests all 429ed is not retried against the same host and is reported as rate-limited rather than `no-results` when filtering leaves it empty.diff --git a/changelog.d/+reddit-run-memo.fixed.md b/changelog.d/+reddit-run-memo.fixed.mddeleted file mode 100644index 5d8516d1..00000000--- a/changelog.d/+reddit-run-memo.fixed.md+++ /dev/null@@ -1 +0,0 @@-Keyless Reddit GETs are memoized for the life of one command, and concurrent requesters for the same URL share the in-flight fetch. Subreddit listing partials, listing RSS feeds, arctic supplements, and shreddit comment pages were fetched once per subquery (four times on a typical run) because the Reddit lane is dispatched with the raw topic every time; a four-subquery run now issues roughly 50 reddit.com requests instead of ~184, which is what kept tripping the anonymous rate limit.diff --git a/changelog.d/+source-outcome-swallowed-lane-failure.fixed.md b/changelog.d/+source-outcome-swallowed-lane-failure.fixed.mddeleted file mode 100644index e0f42a5b..00000000--- a/changelog.d/+source-outcome-swallowed-lane-failure.fixed.md+++ /dev/null@@ -1 +0,0 @@-Sources that return items are no longer branded `auth-failed` / `partial` when a swallowed lane-level HTTP failure (e.g. Reddit shreddit partials 403-ing on datacenter egress) is captured by the pipeline sink. The transport-failure outcome still surfaces when nothing was delivered, so `doctor` prescriptions are unaffected.diff --git a/changelog.d/+top-comments-all-clusters.fixed.md b/changelog.d/+top-comments-all-clusters.fixed.mddeleted file mode 100644index eae07dff..00000000--- a/changelog.d/+top-comments-all-clusters.fixed.md+++ /dev/null@@ -1 +0,0 @@-`## Top Community Comments` and `## Best Takes` now draw from every cluster that clears the relevance floor, not only the eight clusters shown in `## Ranked Evidence Clusters`, so a top-voted comment on a lower-ranked thread still reaches the brief.diff --git a/changelog.d/906.fixed.md b/changelog.d/906.fixed.mddeleted file mode 100644index 619d99fd..00000000--- a/changelog.d/906.fixed.md+++ /dev/null@@ -1 +0,0 @@-Keyless Reddit comment-enrichment slots now go to the most-commented threads within each entity-priority tier, so a high-discussion thread no longer misses `## Top Community Comments` coverage while near-empty threads consume the scarce slots.diff --git a/changelog.d/985.fixed.md b/changelog.d/985.fixed.mddeleted file mode 100644index 2679a776..00000000--- a/changelog.d/985.fixed.md+++ /dev/null@@ -1 +0,0 @@-Keyless Reddit now paces unauthenticated reddit.com requests at 1 req/sec (configurable via `LAST30DAYS_REDDIT_KEYLESS_RATE`) and retries a 429'd RSS or listing sub-request once, so routine runs no longer drop those lanes as `partial` HTTP 429.diff --git a/changelog.d/990.added.md b/changelog.d/990.added.mddeleted file mode 100644index 37f0fbc3..00000000--- a/changelog.d/990.added.md+++ /dev/null@@ -1 +0,0 @@-Telegram public channel source: opt-in via `--telegram-sources=handle1,handle2` or `TELEGRAM_SOURCES` + `INCLUDE_SOURCES=telegram`. Named public channels only (no keyword discovery); fetches recent posts via ScrapeCreators API and scores by views, reactions, and topic relevance.diff --git a/gemini-extension.json b/gemini-extension.jsonindex d339df35..898abe17 100644--- a/gemini-extension.json+++ b/gemini-extension.json@@ -1,6 +1,6 @@{"name": "last30days-skill",- "version": "3.21.1",+ "version": "3.22.0","description": "Research a topic from the last 30 days across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, and the web.","settings": [{diff --git a/pyproject.toml b/pyproject.tomlindex d3264ace..c578707c 100644--- a/pyproject.toml+++ b/pyproject.toml@@ -1,6 +1,6 @@[project]name = "last30days-skill"-version = "3.21.1"+version = "3.22.0"description = "Multi-source last-30-days research skill"readme = "README.md"requires-python = ">=3.12"diff --git a/skills/last30days/SKILL.md b/skills/last30days/SKILL.mdindex e21d8567..b683f051 100644--- a/skills/last30days/SKILL.md+++ b/skills/last30days/SKILL.md@@ -1,6 +1,6 @@---name: last30days-version: "3.21.1"+version: "3.22.0"description: "Research what people actually say about any topic in the last 30 days. Pulls posts and engagement from Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, and the web. Includes a doctor health check to diagnose broken or missing sources."argument-hint: 'last30days nvidia earnings reaction | last30days AI video tools | last30days what users want in react'allowed-tools: Bash, Read, Write, AskUserQuestion, WebSearch@@ -412,7 +412,7 @@ If your Bash call to `last30days.py` does NOT include the FULL pre-flight checkl----# last30days v3.21.1: Research Any Topic from the Last 30 Days+# last30days v3.22.0: Research Any Topic from the Last 30 Days> **Permissions overview:** Reads public web/platform data and optionally saves research briefings to `LAST30DAYS_MEMORY_DIR` (defaults to `~/Documents/Last30Days`). X/Twitter search uses optional user-provided tokens (AUTH_TOKEN/CT0 env vars). Bluesky search uses optional app password (BSKY_HANDLE/BSKY_APP_PASSWORD env vars - create at bsky.app/settings/app-passwords). On hosts with `uv` and no Python 3.12+, the preflight may install a uv-managed CPython 3.12 (one-time ~28MB download, announced on stderr). All credential usage and data writes are documented in the [Security & Permissions](#security--permissions) section.diff --git a/uv.lock b/uv.lockindex 45e0707b..39fea5bd 100644--- a/uv.lock+++ b/uv.lock@@ -130,7 +130,7 @@ wheels = [[[package]]name = "last30days-skill"-version = "3.21.1"+version = "3.22.0"source = { virtual = "." }[package.dev-dependencies]