docs: redesign the README as a landing page with an inbox-triage example - #227
Merged
Merged
Conversation
- Centered header with a new light/dark icon, tagline, badges and links. - Lead with the compact `caliper compare` screenshot, then a feature grid and a side-by-side quick start (agent path / CLI path). - Replace the commit-writer example with inbox-triage vs calendar-scheduler: six tasks covering the LLM judge, Python assertions, a prompt injection, a hijacked neighbour prompt and a silence probe. The ablation is realistic: the bare agent passes 80%, the skill 100% with 30% fewer tokens. - Add a Documentation index and a License section; the reference sections are unchanged. - docs/render_readme_samples.py renders the new example, so the SVGs stay reproducible. - docs/results.md now carries the "one attempt = one agent session" cost note that the landing page drops.
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
Point both paths at inbox-triage/inbox-triage.eval.yaml, where grill-skill writes the spec. Run the ablated control before the full run, so the bare spec name resolves to the full run in caliper compare. Replace the <ablated-run> <full-run> placeholders, which the shell reads as redirections, with a results path and the spec name.
Drop the "Then:" prose and the redundant --k 3 (3 is the default), and name the spec path once with SPEC=.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Rework
README.mdinto a landing page.Before / after
The first screen of the README, rendered with GitHub's styles:
Changes
docs/assets/icon-{light,dark}.svg), the tagline, the "next nine times?" hook, badges, and quick links.caliper comparescreenshot comes first. The longcaliper runoutput moves further down, to a "Read the output" section after the spec.inbox-triagevscalendar-scheduler, replacingcommit-writer. Its six tasks cover:expect:)activates: [])Supporting changes
docs/render_readme_samples.pynow renders the inbox example, socompare-ablation.svgandrun-output.svgcan be regenerated withpython docs/render_readme_samples.py.compare-runs.svg, used bydocs/results.md, regenerates byte-identical.docs/results.mdnow has the "each attempt is one agent session" cost note that the landing page drops.No CLI flag, spec format, judge behavior, or results schema changes, so no other docs from the CLAUDE.md list need updating. Every
README.md#…anchor linked fromdocs/andexamples/still resolves.The before/after screenshots live in commit
8c540d2, which the next commit removes, so they don't add to this PR's diff.Checks
ruff format --check .andruff check .pass.caliper validate. Its setup commands run and itsassert:blocks compile.pytestwas not run locally; this PR only touches docs and the docs render script, and CI runs the suite.The screenshot numbers come from sample data in the render script, not a real run, as before.