AgentCortex 1.0.1

Removed (agent-exclusive files cleanup: CLAUDE.md and AGENTS.md symlink)

• Why: Global rules were revised on 2026-10-12 (by the user, superseding the previous "sub-project .claude/ superset" rule): agent-exclusive files — CLAUDE.md, AGENTS.md, .claude/, project-level .pi/, and other project-level files that are loaded at session start from the working directory and represent agent identity — may only be owned by Agent projects; non-Agent projects (including sub-projects) must not contain them, and any that remain are cleaned up on sight.
• What: Deleted the project-root CLAUDE.md and the AGENTS.md symlink (their content was the AgentCortex project guide plus an attached full copy of the parent NeuralCoreAgent (Ada) rules). Guide sections duplicated by the README were dropped; the one unique item, the evaluation-documentation entry point, was migrated into the README Project Files table (new docs/ row, synced across EN/ZH). The attached Ada rules belong to the NeuralCoreAgent project, so the copy here was deleted. The repository now contains no agent-exclusive files.

Added (AGENTS.md symlink pointing to the project-root CLAUDE.md)

• Why: CLAUDE.md had already moved from .claude/ to the project root (following the global convention "keep the persona CLAUDE.md at the project root and add an AGENTS.md symlink pointing to it"), so the symlink lets agent tools that only recognize AGENTS.md (e.g. ZCode) load the same project guide; the two files stay in sync naturally with no dual maintenance.
• What: Created the project-root symlink AGENTS.md → CLAUDE.md (ln -s CLAUDE.md AGENTS.md); CLAUDE.md itself was not modified.

Changed (migration wrap-up: update the sub-project list path in CLAUDE.md)

• Why: The project now resides under ~/Developer/ (the former ~/Documents/Projects/ location was retired; during the 2026-09-08 migration wrap-up, the parent NeuralCoreAgent sub-project list and this repo's .claude/CLAUDE.md still pointed to the old path), preventing future sessions from being led to a nonexistent location.
• What: In .claude/CLAUDE.md, the self path referenced was updated from ~/Documents/Projects/AgentCortex to ~/Developer/AgentCortex.

Changed (rename Visitors badge to Visits/day (14d): align alt text with the xhqing centralized stats label)

• Why: Per the user's request (2026-08-17), the visitor badge name should express the "average daily visits over the recent two weeks"; the badge JSON label maintained by the xhqing centralized stats was changed from Visitors to Visits/day (14d) (Visits/day is the conventional shields.io notation for a daily average, and (14d) marks the 14-day rolling window). README alt texts across repos were aligned so the alt no longer drifts from the text actually displayed on the badge.
• What: In the README badge area, alt="Visitors" → alt="Visits/day (14d)"; only the alt text changed — the endpoint URL, data source, and badge semantics are unchanged (the semantics change is recorded in the xhqing repository's CHANGELOG; this repo only updates the alt text).

Changed (capitalize the Visitors badge alt text: unify README visitor-badge naming)

• Why: Per the user's instruction (2026-08-16) to "unify the Visitors badge globally, with a capitalized first letter", and the new global rule that badge English text must be capitalized: the visitor badge added when centralized stats launched used the lowercase alt="visitors", inconsistent with the badge JSON label (Visits/day) and the capitalization convention; this change closes that gap.
• What: In the README (EN/CN) badge area, the visitors badge changed from alt="visitors" to alt="Visitors"; only the displayed alt text changed — the endpoint URL and data source are unchanged.

Added (README visitor badge — team-wide centralized traffic statistics)

• Why: The team rolled out centralized, truly deduplicated traffic statistics (image-badge solutions cannot deduplicate, so this uses the official Traffic API): the stats service is hosted centrally in the xhqing repository (scripts/update_traffic.py plus a daily GitHub Action), and each team repository only needs to show the badge in its README with zero runtime burden.
• What: Added a visitors badge to the README (EN/CN) badge area (a shields.io endpoint pointing to traffic/badges/<repo>.json in the xhqing/xhqing repository, updated from daily-collected official Traffic API data). Meaning of the number: cumulative daily-deduplicated visitors (GitHub only exposes per-day uniques, with no deduplication across days), accumulated since 2026-08-16.

Changed

• Changed project license from PolyForm Noncommercial License 1.0.0 to MIT License: open up the project for commercial use and redistribution, making it easier to promote and adopt. Updated LICENSE.md to the MIT License text; synced the license badge, file table, and LICENSE section in README.md and README_cn.md; simplified the contribution licensing section in CONTRIBUTING.md and CONTRIBUTING_cn.md (the noncommercial restrictions, patent clause, and extra commercial grant to maintainers are no longer needed under MIT)
• Enhanced the LLM-as-Judge evaluation plan (.trae/documents/evaluation-plan/llm-as-judge-evaluation-plan.md) to make the evaluation of the Rapid-Reasoning Engine statistically sound and budgeted: added an evaluation condition baseline enforcing the same "max reasoning effort" (reasoning_effort=max, temperature=1.0, top_p=0.95) for both experimental and control groups so the engine rule is the only variable (motivated by DeepSeek-V4-Flash-0731's built-in max-level deep reasoning, which would otherwise confound the attribution); added sample-size planning (12/30/60 tiers with Cohen's dz power analysis and Wilcoxon/t-test decision criteria); added length-bias control (answer_tokens covariate, explicit judge rule, and matched-length robustness check) since the engine's exhaustive-extraction rules systematically lengthen answers; added a budget estimation section based on official DeepSeek-V4-Flash-0731 API pricing (input $0.14/M cache-miss, output $0.28/M), estimating ~$0.03 for a minimal 12-question run, ~$0.25 for the recommended 30-question ×3-averaged run, and up to ~$0.95 for a fully robust 60-question ×3×3 run with one retest round — concluding token cost is not the bottleneck, the recommended tier can detect a 1-point/10 improvement

AgentCortex 1.0.0

Added

• Tool calling rules V0 (English and Chinese)
• Tool calling rules V1 (English and Chinese)
• DeepSeek V4 Pro Max Thinking rules (Infinite-Reasoning Engine, English and Chinese)
• DeepSeek V4 Flash Max Thinking rules (Rapid-Reasoning Engine, English and Chinese)
• GLM 5.1 Max Thinking rules (Incisive-Reasoning Engine, English and Chinese)
• CONTRIBUTING.md and CONTRIBUTING_cn.md guidelines
• README.md and README_cn.md project documentation
• LICENSE.md (MIT License)
• .gitignore (exclude monetization-plan.md, an internal business plan that must not be committed to the public repository; also exclude tmp/ and .DS_Store)

Changed

• Expanded project scope from tool calling rules to a comprehensive agent rules collection
• Enhanced tool calling rules from V0 to V1 with systematic improvements
• Updated documentation with detailed comparison of three reasoning engines
• Refreshed CONTRIBUTING, LICENSE, and README files