Popular repositories Loading

1. filename-as-instruction filename-as-instruction Public

   Put the reading constraint in the filename, not in the file. A one-step technique for stopping AI coding agents from partially reading documents that must be read in full.

2. who-said-what who-said-what Public

   When one agent summarises several agents, it gets the attributions wrong - and it is the only participant who cannot notice. A one-round-trip check.

3. rehearsal-blind-spots rehearsal-blind-spots Public

   A dry run with subagents cannot surface the failures that only appear when something stops. One question decides which runs must be real.

4. the-record-instead-of-the-act the-record-instead-of-the-act Public

   An agent can carry a task as far as writing it down - and the writing down becomes the doing. Not fabricated tool output: the record is genuine, the act is missing.

5. caas caas Public

   CAAS (Chat–Agent Auto Session): talk only to your everyday Claude.ai conversation; running Claude Code sessions do the work and wake the same conversation with the result. Includes a tool-limiting …

   JavaScript

6. terminal-japanese-input-pad terminal-japanese-input-pad Public

   Always-on Japanese input box docked under Windows Terminal (avoids IME candidate window overlap)

   Python