Repository navigation
New thought: when a small decision model can replace an LLM call - #29
Open
Tim-Conrad wants to merge 10 commits into
Open
Tim-Conrad wants to merge 10 commits into
Tim-Conrad wants to merge 10 commits into
Conversation
Starts from the Strands Agents post on Amazon replacing Sonnet 4.6 with Strands Decider for ticket labels. Draft. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
Rewritten around the method from the repository's customize-decider guide: task card, frozen test set, baselines, the ladder from rewording to training, thresholds, a side-by-side run, and watching. Adds a stepped diagram of the eight steps. A stepped figure can now hold a narrow drawing for phones, as an animated one can. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
How sealed works, concepts first: what the host sees, from password to file, the service worker, keyed blob names, groups and new keys, a comparison with StatiCrypt, and the limits. Four figures. Draft. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
Owner
Author
Handover: the steps that are still openThis PR now holds three thoughts' worth of work: decision models, sealed, and the link to the sealed demo. It depends on work in a second repository, Tim-Conrad/sealed, branch State
What
Steps1. sealed: open a PR and merge it
2. sealed in the lab: deploy and check
3. The deploy key for the demo
4. Publish the demo and check it
5. Ship this PR, only after step 4
6. Later: before sealed goes public
Generated by Claude Code |
DPoP-bound tokens, a local broker that identifies the calling program (pidfd, binary and script hash, systemd unit, as macOS does with code signatures), a phone app that shows the intent and signs a short grant bound to the agent's key, and agentgateway as the place for the last upstream keys. Five figures, a prototype plan and the limits. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
The game pages get their own section: the characters are the game's own art (Wortzauber sprites, Kleinbahn cars rendered from the near models), everything around them is drawn in ink, they stand on the sheet's lines, move on their own and answer taps and drags, stand still for reduced motion, and stay under about 400 KB per theme. The motion rule names them as an exception, a game page drops the screenshot grid, and the component table and file list name the new classes and files. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
…one loop Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
…site-redesign-oiralc # Conflicts: # index.html
On the Wortzauber page Paul and the Schwebe-Bot walk on the title, the rules, the feature list and the buttons. You can pick Paul up and throw him, and dotted words in the text make their thing appear. He follows the reader down the page. On the Kleinbahn page the rule under the header is a railway with two tunnels that form one loop, a station, a signal and sheep. The train is rendered from the game's near models, by day and by night, with turning wheels and low-poly smoke. Tap it to stop or start it, push it, or set the signal. Both pages drop the screenshot grid; the videos show more. The front page is unchanged. CLAUDE.md names the two new scripts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KFxuEuiQ8gLmh8xHfJyLFW
This branch has not been deployed
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
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.
A new thought at
/thoughts/decision-models/, status Draft. It starts from the Strands Agents post on Amazon replacing Claude Sonnet 4.6 with Strands Decider for ticket labels, and focuses on the method: how to do this well, at home and at work.What changed
thoughts/decision-models/index.html: the thought.customize-deciderguide: fit, task card, frozen test set, baselines, the ladder from rewording to training, thresholds, a side-by-side run, and watching.thoughts/index.html,index.html,sitemap.xml:assets/site.js,assets/site.css,DESIGN.md: a stepped figure can now hold a narrow drawing for phones (.dg-n), as an animated one can.data-stepparts in the whole figure, not only in its first SVG.Sources
Checks
python3 tools/check.py: 12 pages checked, 0 problems.Generated by Claude Code