docs: remove remaining em dashes in UI configuration notes

docs: clarify Authorship & Contributors section for Gabriel Devenyi

release: v22.4 - Multi-tool token alignment, extended reasoning formats, and UI default knobs

docs: update authorship credits in README

fix(v22.3): KV cache invariance, script reorganization, and v22.2 archiving

Merge PR #84 from g-a-b-y: v22.3 error detection tiering, reasoning fixes, test suite 44->101 + property fuzzer

v22.3: Error detection tiering, reasoning fixes, test suite 44→101 + property fuzzer

chore: remove .gitignore in favor of local git exclude

Delete .agents

release: v22.2 - Extended effort aliases, tool error disambiguation, and multi-system prompt merging

chore: persist workspace rules for style invariants and Qwen template architecture

release: v22.1 - Safe medium reasoning default, inline chat tags, and check_applied diagnostic

release: v22 - Add Qwen 3.8 support, reasoning effort steering, and engine fixes

docs: Update 'Restored' to 'Maintained' for XML tool format fix in README

docs: Add v20 and v21 fixes to the Technical Details section

docs: Simplify Critical Issues Fixed table and add missing v20/v21 features

docs: Collapse Quick Install by default and remove obsolete comparison matrix

docs: Add smart collapsible sections to README for better readability

docs: Complete restructuring and humanization of README to prioritize quick install and core features

docs: Rewrite JSON kwarg section for better clarity and less confusion

docs: Clarify that mainline llama-server supports XML and the JSON kwarg is strictly for outdated/custom Hermes parsers

docs: Add dedicated README section for JSON format override kwarg

v21.3: Add optional JSON tool format kwarg override

v21.2: Fix reasoning bypass hallucination by adjusting <IMPORTANT> block

Revert PR 45 to restore native XML compatibility

Fix non-thinking generation prompt formatting

Bump version to v21 and archive v20

Merge PR 50

Merge PR 47

Merge PR 48

Merge PR 49

Merge PR 45

Fix: preserve_thinking default regressed to false in v20 (contradicts v19 changelog & README)

Remove extra newline that breaks prefix cache after tool calls (#42)

Enhancement: support Anthropic-style `message.thinking` alongside `reasoning_content`

Fix: <|think_on|>/<|think_off|> honored from tool responses (prompt-injection surface)

Fix: inline-quoted </think> in assistant history swallows half the message as reasoning

Fix: false-positive tool-error detection (marker anywhere in response) + add npm/git markers

Fix: tool calls silently lost on llama.cpp/ik_llama — switch XML format to Hermes JSON

docs: update changelog and details with minja replace filter hotfix

fix: bypass minja C++ replace filter bug dropping text payload at index 0

feat: Upgrade to v20 (The Architect Patch) with C++ AST optimization and deep agent loop fallbacks

Release v19: Cured 'Empty Think' poisoning and logic traps, enabled chronological reasoning preservation by default

Release v18: Improved error detection, engine compatibility, and whitespace normalization

Archive v17 template and readme

v17 release: Unified template, 100% KV Cache hit rate optimization, and Stopping Bug fix

Make update history collapsible in README

Initial upload: Qwen-Fixed-Chat-Templates