Moore2877/Qwen-Fixed-Chat-Templates-llamacpp
chore: retire v20 root copies (archived under archive/)
v20.1: cross-harness skill loading, delivery reserve, correctness fixes (local eb8ffcb)
Release v20: action-staged skills and bounded todo maintenance
chore: retire v18 root copies (archived) and remove stray __pycache__ bytecode
v19: history-rendering correctness and official-template alignment (local e1be1be)
v18: bound reasoning effort and investigation
v17: retire root v16 artifacts (archived)
v17: retire root v16 artifacts (archived)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
v17: Qwen 3.8 support (reasoning_effort, preserve_reasoning, system merging)
release: remove stale root v15 aliases after v16 promotion
v16: cumulative agentic reliability and skill/delegation release
v15: mid-context skill rechecks and versioned artifacts
v14: skill triggering - final artifacts (oneline, v13 opt-out + archive, tests)
v14: skill triggering - final artifacts (oneline, v13 opt-out + archive, tests)
v14: skill triggering - final artifacts (oneline, v13 opt-out + archive, tests)
v14: skill triggering - final artifacts (oneline, v13 opt-out + archive, tests)
v14: skill triggering - final artifacts (oneline, v13 opt-out + archive, tests)
v14: skill triggering - final artifacts (oneline, v13 opt-out + archive, tests)
v14: skill triggering (frontier-parity skill loading; v13 shipped alongside for opt-out)
v14: skill triggering (frontier-parity skill loading; v13 shipped alongside for opt-out)
v14: skill triggering (frontier-parity skill loading; v13 shipped alongside for opt-out)
v13: resolve tool results by tool_call_id
v12: two-sided turn management
v11: language release — research-driven rewording of all model-visible phrases
tidy: remove stray blank lines left by comment stripping (render-identical; oneline unchanged)
v10: truthfulness release + comment-free template
v9: ping-pong/action-loop detector + vanished-tools note
docs: note JSON mode drops grammar enforcement (verified: tool_choice=required unconstrained in JSON)
docs: add a fully-loaded --chat-template-kwargs example (one JSON object; PowerShell escaping note)
docs: slim down README (one-line changelog, tighter notes); credit AI assistant; extend archive table to v7
v8: media-marker correctness — strip only unambiguous roles (system/assistant/reasoning), preserve user/tool markers (fixes real-vision & tool-screenshot desync from v3/v7 unconditional strip); archive v7
docs: strip-proof media-marker reviewer notes (in-template + README) so fetch tools that eat the <...> token don't misread the guard as an empty-string bug
v7: runtime-hardening (repeat_nudge=0 fix, whitespace fidelity, long-error detection, head+tail truncation) + repro script + expanded tests; archive v6
archive: preserve v1-v5 template versions with lineage README
v6: cross-harness mutating_tools union (16+ harnesses) + case-insensitive match
v5: remove same-tool-streak detector (90% noise), add mutating-tool exclusion to no-progress detector (data-grounded on 381 sessions)
docs: set kwargs server-side (--chat-template-kwargs); clients may drop them
v4: tool-envelope unwrapping, self-healing spilled tool calls, think_on_tool_failure kwarg, emoji-free warnings
v3: media-marker injection guard (ik_llama tokenize-poisoning fix)
v2: strftime_now date awareness (llama.cpp-native, guarded) + verified llama.cpp integration notes
tests: resolve template path relative to repo root
v1: froggeric v21.3 base + agentic loop detection, harness-aware error markers, threshold kwargs
initial commit