hablas-ai 2.1.0 → 2.1.2

package/CHANGELOG.md

@@ -1,29 +1,25 @@
 # Changelog
 
-## [2.1.0] — 2026-06-12 · Apple UX + Tool Hardening
+## [2.1.2] — 2026-06-12 · Full-Read Context Fix
+
+### Fixed
+- `read_file` results are now preserved in full inside the model context instead of being aggressively truncated.
+- This prevents the agent from rereading the same large file just because it only saw the first chunk in context.
+- Tool result boxes now summarize the first meaningful line instead of flattening huge file bodies into unreadable one-liners.
+
+## [2.1.1] — 2026-06-12 · Thinking, Streaming, and Read-Dedupe UX Fix
 
 ### Added
-- restored advanced shell UX:
-  - tab completion for slash commands
-  - tab completion for common file paths
-  - `#file` completion
-  - command history persistence
-- Apple-slate inspired terminal UI rebuild:
-  - richer banner
-  - improved assistant rendering
-  - improved tool cards
-  - cleaner help and status formatting
-- `download_asset` tool for bringing real remote assets into the project
-- stronger web/image workflow in the system prompt
-- stronger parser-side sanitization for raw tool markup leakage
+- visible thinking state while Hablas is waiting on the model
+- real streaming for direct text-only turns
+- stronger shell completion and history workflow kept active
 
 ### Changed
-- rebuilt single-agent system prompt using the strongest parts of the old specialist prompts
-- strengthened implementation enforcement so real engineering work is expected to use tools
-- improved implementation classification, including typo handling and image/wallpaper related requests
-- improved read/analysis/implementation rendering paths in the terminal
+- tool cards were rebuilt again for cleaner boxed output
+- assistant rendering is more consistent between direct and tool-driven turns
+- the prompt was tightened for shorter, more natural casual and identity responses
 
 ### Fixed
-- raw XML-like tool leakage is sanitized more aggressively
-- implementation prompts that try to answer with inline code only are pushed back toward real tool execution
-- setup flow remains restored with the classic two-choice onboarding model
+- repeated `read_file` calls for the same path in the same turn are now deduplicated and reused from cache unless the file changed
+- raw tool-markup leakage is sanitized more aggressively
+- implementation and image/wallpaper requests remain wired to the stronger tool path

package/README.md

@@ -100,6 +100,7 @@ This means Hablas can:
 - search the web for references and assets
 - inspect pages and extract links
 - download real images/files into the project when appropriate
+- preserve full file reads in-context so long files are not silently cut into tiny fragments that force rereads
 
 ---