db8fb41764
license: MIT LICENSE in root, plugins, and all skills
2026-04-05 19:35:33 +01:00
3c37563f0c
v3.4.28: debate skill — evidence-based multi-agent debates with Discovery phase
2026-04-05 18:21:42 +01:00
6d2139cc05
v3.4.27: glm-design-to-code auto-fix pipeline
...
- Phase 3 Step 1: extract validation with exit codes + manual fallback
- Phase 3 Step 3: auto-fix loop (max 3 attempts, compilation-only fixes)
- Phase 4: mandatory verify with framework-specific timeouts + blank detection
- Mode routing and error handling table updated
2026-04-05 15:36:43 +01:00
e50cf7b8c2
v3.4.26: trial skill + gitignore d2c-output
...
- Add glm-design-to-code-trial standalone skill for npx skills distribution
- Add brewpage-reference.png test asset
- Gitignore d2c-output/ (generated artifacts)
2026-04-04 07:57:59 +01:00
534e4567a0
v3.4.25: glm-design-to-code smart intent detection + dual HTML input
...
- Smart intent classification (reproduce/creative/enhance/modify/convert)
- Dual input for HTML: screenshot + HTML source sent together
- Custom instruction param (GLM_INSTRUCTION) replaces hardcoded prompts
- Profile prompts decoupled from intent (quality-only)
- ARG_MAX safety: rawfile for user_text in both scripts
- Routing table: 5-way payload dispatch by input type
2026-04-04 07:47:46 +01:00
74b39e7866
v3.4.24: glm-design-to-code auto-invocation + mandatory config output
...
- Removed disable-model-invocation — auto-triggers on natural phrases
- Optimized description: 3 natural examples, "external GLM API" signal
- Mandatory resolved configuration table before every API call
- Parameter priority: prompt flags > inline text > .env > defaults
- --model auto-prefix z-ai/ for OpenRouter, auto-strip for Z.ai
- Raw text scanning for inline API keys and model names
2026-04-03 21:21:26 +01:00
b8cc538186
v3.4.23: glm-design-to-code review fixes + docs enhancement
...
Expert panel review (6 agents) → 11 confirmed fixes:
- Context window 128K→202K, phantom model fix, pricing standardization
- API key flow: validate before save, split provider/key questions
- Scripts: +x permissions, curl 600s timeout, OpenRouter headers
- glm-extract.sh: allow spaces in paths, quote awk system() args
- glm-verify.sh: track and cleanup background timer PID
- REVIEW mode: use parsed RESULT_IMAGE instead of hardcoded path
Docs: benchmark comparison, research results tabs, external resources
2026-04-03 20:38:55 +01:00
592de4e93b
v3.4.22: add glm-design-to-code skill
...
New skill: /brewcode:glm-design-to-code — GLM vision model-powered
design-to-code generator with 4 input types (screenshots, text
descriptions, HTML files, URLs), 3 modes (CREATE/REVIEW/FIX),
3 frameworks (HTML/CSS, React 18, Flutter Web), 3 quality profiles,
dual provider support (Z.ai + OpenRouter).
Includes: SKILL.md, 6 pipeline scripts, 7 reference prompts,
2 specialist agents, full docs site page with benchmarks,
plugin README/commands updates, navigation integration.
2026-04-03 18:13:57 +01:00
34d0eb4af5
v3.4.21: add brewdoc:guide docs page
2026-04-03 08:02:02 +01:00
a37064f24d
v3.4.20: add brewdoc:guide — interactive teaching skill
2026-04-03 07:56:18 +01:00
66638e9c69
v3.4.19: add brewcode:e2e skill — E2E testing orchestration
...
6 modes (setup/create/update/review/rules/status), 5 runtime agents,
BDD scenarios with YAML frontmatter, layered test architecture,
quorum review (3 reviewers, 2/3 consensus), MAX_CYCLES=3 review loops.
Stack-agnostic with mapping for Java/Python/JS/C#.
2026-04-02 20:08:53 +01:00
554c722056
v3.4.18: fix version check and config merge
...
- parseVersion: fallback to plugin.json when path has no version
- setup.sh: merge config instead of overwriting (preserves user settings)
- version check logs use [version] prefix
2026-04-02 17:53:27 +01:00
82c7f634f7
v3.4.17: fix brewpage-publish W007 insecure credential handling
...
ownerToken no longer appears in conversation output. Curl response
parsing and history saving now execute atomically inside bash.
LLM only sees the published URL.
2026-04-02 17:10:21 +01:00
2aeff9a6d5
v3.4.16: merge install skill into setup Phase 0
...
Prerequisites check (brew, coreutils, jq, grepai) now runs automatically
as Phase 0 before project analysis. If all required tools are present,
Phase 0 is skipped silently.
2026-04-02 14:08:47 +01:00
b3fe7aa269
v3.4.15: teams session-scoped trace system (trace.jsonl replaces 3 MD files)
2026-04-02 11:34:15 +01:00
65c0861013
fix(brewtools): remove duplicate hooks reference from plugin.json
2026-04-02 09:09:49 +01:00
5a6a0510c0
v3.4.14: brewtools plugin, teams skill, mode switcher, hook-creator v2.1.89+
...
brewtools (new plugin):
- text-optimize, text-human, secrets-scan skills + text-optimizer agent
- BT_PLUGIN_ROOT injection via session-start hook
brewcode:
- /brewcode:teams — dynamic agent team creation/management
- Mode Switcher system (modes/, getActiveMode(), hooks injection)
- plan: Dynamic Agent Resolution from .claude/teams/ roster
- skills: Step 2.5 mode-switcher detection
- skill-creator: Mode Switcher pattern docs
- hook-creator: v2.1.89+ (PermissionDenied event, defer, 26 events)
- removed: text-optimize, text-human, secrets-scan → brewtools
- convention P5: fallback when brewtools not installed
2026-04-02 08:22:13 +01:00
kochetkov-ma