Releases
Every release entry is generated at build time from the canonical release feed, then published with a matching downloadable ZIP.
Changes since v2.9.0
- Tighten instruction-evaluator and acceptance-tester review criteria (08d6faf)
Changes since v2.8.0
- split agent-vs-skill definition (aa170b8)
Changes since v2.7.2
- introduce test system kit (4ff310e)
Changes since v2.7.1
- introduce a new agent that runs scenario-based acceptance tests against new or materially changed instruction artifacts before they are accepted.
Changes since v2.7.0
- fixe release pipeline packaging/trigger config: add missed conventions and agents subfolders to final release files (60767ee)
Changes since v2.6.0
- Traceability now applies to non-trivial routed handoffs, not every gated capability or internal command.
- Manager no longer must “re-read this file after every step.” It now checks artifacts at routed handoff points and reloads instructions only when the plan is ambiguous, changed, or missing from context.
- Task-complete now references artifact labels or transcript locations when available, instead of requiring brittle message citations. Review/evaluator checks now look specifically for missing artifacts in non-trivial routed work
Changes since v2.5.1
- The evaluator now has a task-agnostic Substantive Coverage review area that checks whether an artifact actually covers the responsibility implied by its name, description, triggers, inputs, and output contract. It also adds skill-specific coverage checks so a structurally valid skill can still be flagged when its procedure/checklist undercovers its declared domain. (81db345)
Changes since v2.1.2
- 2.5.1: Added the explicit convention reference to protocols/documentation_maintenance.md. (a1b52cf)
- 2.5.0: New protocols/documentation_maintenance.md mandatory protocol. (652f15f)
- 2.4.0: Changed IMPLEMENTATION.md to require selective-loading structure for reference docs and apply the new convention. (d3182ef)
- 2.3.0: Introduce convensions/*.md as a framework source to replace repeated prose in project instructions (625f618)
Changes since v2.1.1
- Clarify protocol explicitness rules (0a7a098)
Changes since v2.1.0
- Implement first mandatory Agent: instruction-evaluator (58c89af)
Changes since v2.0.0
- reduce number of principles, make the system more clear and easy to remember (4ebd124)
Changes since v1.5.4
- refactor: Update documentation to clarify protocol adherence and layer purity standards (9a410e4)
- tool-specific adapters must be explicit mandatory shims (0648235)
- refactor: Restructure documentation to enhance clarity and compliance with implementation principles (d5f9e73)
- refactor: Clarify convention guidelines by specifying that they should not classify, route, sequence, or execute work (b0f7967)
- refactor: Replace 'rules' with 'conventions' throughout documentation for consistency (9810224)
- refactor: Enhance brainstorming protocol and clarify profile capture process (80ba736)
Changes since v1.5.3
- Rework README as a how-to guide (2235fb2)
Changes since v1.5.0
- Correct release version (71b923b)
- simplimy Medium and Large rules (793f348)
- Bump version to 1.5.2 and update documentation for clarity on risky changes (efd5116)
- Update release script (f4f9119)
- Fix evolution prompt scoping for explicit capability requests (a6b98e6)
- add codex to gitignore (105dd1e)
Changes since v1.4.0
- Clarify multi-AI skill format (be17467)
- Fix README typo (22ede61)
- fix: rebuild release 1.4.0 with missing files (41c87e1)
- Build release notes from commit log (74cf0f5)
- Add tool integration guidance and strengthen framework review checks (f633a62)
- update release to have only valuable files (71366bc)
What's Changed
- align framework with protocol-derived standalone skills and pipeline model by @AlexeyPlatkovsky