diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.json index 4f73f6d..0ad5403 100644 --- a/.claude-plugin/plugin.json +++ b/.claude-plugin/plugin.json @@ -1,7 +1,7 @@ { "name": "fetchwright", "icon": "./.claude-plugin/icon.png", - "version": "0.1.0", + "version": "0.2.0", "description": "Read public web pages that refuse scripts (Reddit threads, pages that answer plain HTTP clients with a 403 or a wall) through Gemini's URL context tool: Google reads the page, a standard-library Python CLI checks the retrieval, drops removed-comment placeholders and leaked usernames, and returns the title and text or a thread's post and comments. Never uses Google Search. A setup skill covers the Gemini API key, tiers, quotas and free-tier privacy. Evergreen.", "author": { "name": "Mark Rogers" diff --git a/CHANGELOG.md b/CHANGELOG.md index b02de9e..a90480b 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -2,7 +2,7 @@ Plugin versions, newest first. Skill-level changes are in each skill's `CHANGELOG.md`; the session log is `ai-docs/log.md`. -## Unreleased +## 0.2.0 ยท 2026-10-16 - A 429 that names one model's own quota (the free tier gives `gemini-3.8-flash` 20 requests a day) moves to the next model, which is then used for the rest of the run; the run stops with exit 3 only when every model's quota is spent, or at any other 429. 39 offline tests. diff --git a/README.md b/README.md index c344057..0e81874 100644 --- a/README.md +++ b/README.md @@ -71,15 +71,15 @@ Other agents that read Agent Skills (`SKILL.md`): link or copy the two folders u fetchwright should stay small enough that nobody thinks twice about installing it. `scripts/budget.py` measures each row, and CI fails at red. -| measure | green up to | yellow up to | now (0.1.0) | +| measure | green up to | yellow up to | now (0.2.0) | |---|---:|---:|---:| | longest skill description, characters (loaded in every session) | 1,000 | 1,300 | 873 | | both descriptions, characters | 1,800 | 2,400 | 1,734 | -| largest SKILL.md, tokens | 2,500 | 3,500 | 1,519 | +| largest SKILL.md, tokens | 2,500 | 3,500 | 1,590 | | largest reference file, tokens | 2,000 | 3,000 | 1,292 | -| fetchwright.py, lines | 600 | 900 | 497 | -| installed skills folder, KB | 150 | 400 | 88 | -| tracked repository, KB | 1,024 | 3,072 | 489 | +| fetchwright.py, lines | 600 | 900 | 510 | +| installed skills folder, KB | 150 | 400 | 92 | +| tracked repository, KB | 1,024 | 3,072 | 523 | ## Evergreen diff --git a/ai-docs/log.md b/ai-docs/log.md index 72fc250..72a52e2 100644 --- a/ai-docs/log.md +++ b/ai-docs/log.md @@ -11,3 +11,4 @@ Append-only. One line per operation: `## [YYYY-MM-DD] op | title` where op is on ## [2026-10-16] build | a 429 on one model's quota moves to the next model (PR #2, merged, unreleased) everscout's first live run hit the free tier's 20-a-day limit on gemini-3.8-flash and fetchwright stopped although flash-lite had quota. Mark agreed to fall through on a per-model quota 429. Lessons L-004 and L-005 (PR #1). everscout runs this clone through its `fetchwright.cli` config until a 0.2.0 release, which needs Mark's go-ahead. +## [2026-10-16] release | fetchwright 0.2.0: the quota fall-through (PR #2), every version field bumped, README budget table refreshed (all green); released on the owner's go-ahead diff --git a/plugin.json b/plugin.json index 3c70567..45abcb5 100644 --- a/plugin.json +++ b/plugin.json @@ -1,6 +1,6 @@ { "name": "fetchwright", - "version": "0.1.0", + "version": "0.2.0", "description": "Read public web pages that refuse scripts (Reddit threads, pages that answer plain HTTP clients with a 403 or a wall) through Gemini's URL context tool: Google reads the page, a standard-library Python CLI checks the retrieval, drops removed-comment placeholders and leaked usernames, and returns the title and text or a thread's post and comments. Never uses Google Search. A setup skill covers the Gemini API key, tiers, quotas and free-tier privacy. Evergreen.", "author": "Mark Rogers", "license": "MIT", diff --git a/skills/fetchwright-setup/SKILL.md b/skills/fetchwright-setup/SKILL.md index baeb991..815e540 100644 --- a/skills/fetchwright-setup/SKILL.md +++ b/skills/fetchwright-setup/SKILL.md @@ -2,7 +2,7 @@ name: fetchwright-setup description: "Set up and advise on the Gemini API key that fetchwright needs. Covers creating a key in AI Studio, storing it as GEMINI_API_KEY (setx on Windows, a shell profile on macOS and Linux) without pasting it into chat or a repository, and checking it with fetchwright check. Explains the free and paid tiers, rate limits and 429s, what a read costs, what Google may do with free-tier prompts (used to improve products, read by human reviewers), the 18+ rule, and the EEA, Swiss and UK rule for serving others. Use when the user asks 'set up fetchwright', 'get a gemini api key', 'fetchwright says no key', 'is the gemini free tier private', 'why does gemini give me 429', 'what does fetchwright cost', 'should I pay for gemini', or when fetchwright check exits 4. Also for 'refresh fetchwright-setup'. Not for other Google APIs, Vertex AI, or reading pages (that is fetchwright)." metadata: - version: "0.1.0" + version: "0.2.0" --- # fetchwright-setup diff --git a/skills/fetchwright-setup/evergreen.json b/skills/fetchwright-setup/evergreen.json index 49f855b..3b21899 100644 --- a/skills/fetchwright-setup/evergreen.json +++ b/skills/fetchwright-setup/evergreen.json @@ -38,5 +38,5 @@ "counts": {"learnings": 1, "changes": 1, "research": 2, "tests": 0}, "consolidate_every": 25, "tests": {"last_run": null, "harness": null, "env": null, "cases": 4, "passed": 0, "failed": 0, "failing": [], "last_failure": null, "research_after_days": null}, - "version": "0.1.0" + "version": "0.2.0" } diff --git a/skills/fetchwright/SKILL.md b/skills/fetchwright/SKILL.md index ec15571..afae642 100644 --- a/skills/fetchwright/SKILL.md +++ b/skills/fetchwright/SKILL.md @@ -2,7 +2,7 @@ name: fetchwright description: "Read a public web page the agent's own fetch tools cannot reach (Reddit threads, pages that answer scripts with a 403, a block page or a JavaScript wall) through Gemini's URL context tool: Google reads the page from its index or a live fetch, and a standard-library CLI returns the title and text, or a thread's post and comments, as text or JSON. Needs a Gemini API key in GEMINI_API_KEY (fetchwright-setup sets one up). Use when WebFetch, curl or a plain HTTP request was refused or got a wall on a public page, or the user says 'read this reddit thread', 'get the comments from this thread', 'fetch this through gemini', 'webfetch can't open this page'. Also for 'refresh fetchwright' and 'is fetchwright stale'. Not for pages behind a login or a paywall, YouTube, Google Docs, private or localhost URLs, or web search (fetchwright never uses Google Search)." metadata: - version: "0.1.0" + version: "0.2.0" --- # fetchwright diff --git a/skills/fetchwright/evergreen.json b/skills/fetchwright/evergreen.json index 0b31c47..8d8833c 100644 --- a/skills/fetchwright/evergreen.json +++ b/skills/fetchwright/evergreen.json @@ -64,5 +64,5 @@ "last_failure": null, "research_after_days": null }, - "version": "0.1.0" + "version": "0.2.0" } diff --git a/skills/fetchwright/scripts/fetchwright.py b/skills/fetchwright/scripts/fetchwright.py index 0f7a116..195ef9f 100644 --- a/skills/fetchwright/scripts/fetchwright.py +++ b/skills/fetchwright/scripts/fetchwright.py @@ -26,7 +26,7 @@ import urllib.parse import urllib.request -VERSION = "0.1.0" +VERSION = "0.2.0" API_BASE = "https://generativelanguage.googleapis.com/v1beta" DEFAULT_MODELS = ("gemini-3.8-flash", "gemini-3.5-flash-lite") KEY_VARS = ("GEMINI_API_KEY", "GOOGLE_API_KEY")