Changelog
Unreleased
Section titled “Unreleased”- Project customization (
.maestria/workflow.mdthen.maestria/rules.md) loads as subordinate guidance, re-read everypre_llm_callturn with a visible error banner; see the methodology changelog.
0.1.23 (2026-10-01)
Section titled “0.1.23 (2026-10-01)”/reviewand/plannow reach Hermes’ built-in commands instead of being swallowed by the plugin’s mode switch. The plugin registers/fein,/sonar,/blitz,/mode, and/mode-clear; its reviewer and planner specialists run as/feinpipeline stages.- Sonar mode can now read installed skills and recall past sessions (
skill_view,skills_list,session_search), matching what delegated children already could.skill_managestays blocked, and no write, shell, code-execution, or delegation tool was added to any mode.
0.1.22 (2026-09-30)
Section titled “0.1.22 (2026-09-30)”- Agents reassess explanatory comments and redundant tests before accepting changes, and shared directive text was consolidated with no behavior change; see the methodology changelog.
0.1.21 (2026-09-25)
Section titled “0.1.21 (2026-09-25)”- Agents follow behavior-first testing guidance and deliver work as verified, reviewable slices with designs shown before implementation; see the methodology changelog.
- The opt-in methodology annotation hook and mode footer middleware were removed, and mode-switch pipeline text is unified in one table.
0.1.20 (2026-09-18)
Section titled “0.1.20 (2026-09-18)”- Consistent PR delivery contract with proportional documentation assessment; see the methodology changelog.
0.1.19 (2026-09-17)
Section titled “0.1.19 (2026-09-17)”- Stricter visual delivery evidence contract, with capture, PR-body publication, and readback as distinct stages; see the methodology changelog.
0.1.18 (2026-09-16)
Section titled “0.1.18 (2026-09-16)”- Defined when Sonar research is complete, with Blitz reusing valid context and switching routes for investigation or design work; see the methodology changelog.
0.1.17 (2026-09-11)
Section titled “0.1.17 (2026-09-11)”- Removed dead role and permission scaffolding that ADR-HM-002 already recorded as removed, unified slash-command mode text, and restored the fail-closed denial for unknown mode values; generated skill guidance no longer describes the removed role model.
0.1.16 (2026-09-06)
Section titled “0.1.16 (2026-09-06)”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.15 (2026-09-04)
Section titled “0.1.15 (2026-09-04)”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.14 (2026-09-03)
Section titled “0.1.14 (2026-09-03)”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.13 (2026-08-12)
Section titled “0.1.13 (2026-08-12)”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.12 (2026-08-11)
Section titled “0.1.12 (2026-08-11)”- Cross-cutting agent methodology update; see the methodology changelog.
- Hardened the Hermes adapter by wiring native session and subagent lifecycle trust, immutable mode and child safety allowlists, and the current Hermes plugin registration contracts.
0.1.11 (2026-08-04)
Section titled “0.1.11 (2026-08-04)”- Fixed the Hermes delegation rules adaptation in the synced global rules. The route-scoped rewrite of the canonical Delegation section had left a sync
findAndReplacetransform silently no-oping, so the generatedglobal-rules/SKILL.mdshipped the raw canonical wording instead of the Hermes adaptation. Hermes agents again get the correct guidance: on focused/full routes, use only the 7 maestria specialists and never delegate to Hermes’ built-inexploreorgeneralagents, which bypass the pipeline.
0.1.10 (2026-08-03)
Section titled “0.1.10 (2026-08-03)”Selective routing contract
Section titled “Selective routing contract”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.9 (2026-07-29)
Section titled “0.1.9 (2026-07-29)”Rule-override conditions and communication conventions
Section titled “Rule-override conditions and communication conventions”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.8 (2026-07-28)
Section titled “0.1.8 (2026-07-28)”Blind review and fail-loud iteration exit
Section titled “Blind review and fail-loud iteration exit”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.7 (2026-07-22)
Section titled “0.1.7 (2026-07-22)”Directive recomposition
Section titled “Directive recomposition”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.6 (2026-07-21)
Section titled “0.1.6 (2026-07-21)”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.5 (2026-07-21)
Section titled “0.1.5 (2026-07-21)”Compact agent directives
Section titled “Compact agent directives”- Cross-cutting agent methodology update; see the methodology changelog.
0.1.4 (2026-07-17)
Section titled “0.1.4 (2026-07-17)”Changed
Section titled “Changed”- Orchestrator behavior: Simple tasks now run directly instead of always being delegated to subagents. Complex tasks (multi-file, cross-domain, risky) still use specialist routing. Routine changes are faster with less overhead - no action needed on your part.
0.1.3 (2026-07-17)
Section titled “0.1.3 (2026-07-17)”- pre_gateway_dispatch hook: Intercepts
/fein,/sonar,/blitz,/mode,/review,/plancommands before the agent-busy check - Role-based permission enforcement (historical implementation):
[MAESTRIA_ROLE: <specialist>]indelegate_taskcontext enforced per-specialist tool restrictions - Transform hook: Write operations in fein/blitz mode append methodology annotation to tool results
- Telegram forum topic responses now route to the correct thread instead of General
0.1.2 (2026-07-16)
Section titled “0.1.2 (2026-07-16)”- Plugin loading failure due to
src/layout - added root__init__.pyshim
0.1.1 (2026-07-16)
Section titled “0.1.1 (2026-07-16)”Initial release - same as 0.1.0 (versioning alignment).
0.1.0 (2026-07-16)
Section titled “0.1.0 (2026-07-16)”- Plugin scaffold with
plugin.yaml+register(ctx)entry point - 9 skills: 7 specialists + orchestrator + global rules
- Mode system with file persistence (
/fein,/sonar,/blitz,/mode) pre_llm_callhook for mode injection into user messagepre_tool_callhook for mode-based tool gating (sonar write-block)llm_executionmiddleware (opt-in mode footer viaMAESTRIA_MODE_FOOTER=1)- 7 slash commands:
/fein,/sonar,/blitz,/mode,/mode-clear,/review,/plan - OpenCode CLI routing tool (
opencode_route) for delegating complex coding tasks - Subagent lifecycle hooks (
subagent_start/stopfor pipeline tracking) - Session lifecycle hooks (
on_session_start/end) transform_tool_resulthook (observer-mode write logging)- Distribution via
hermes plugins installfrom this repository (git-based)
Changed (post-release design doc alignment)
Section titled “Changed (post-release design doc alignment)”- Memory engine agnosticism - removed stale Mnemosyne integration claim, removed memory/kanban probes
- Dead code removed:
memory.py(MemoryManager.record() never called) - Cross-platform detection:
shutil.which()replacessubprocess.run(["which", ...]) - Plugin check:
node -e "require(...)"replaces fragile npm filesystem walker