Runtime-core: multi-model tool tolerance, failure guard, rich TUI, subagents, plugins/skills#101

Closed

gnanam1990 wants to merge 36 commits into

mainfrom

zenline-runtime-work

Commits on Jun 6, 2026

Zenline TUI + runtime-core reliability + session file checkpoints

authored and
KRATOS
committed
Slice 3: model registry depth — regex aliases, deprecation fallback, reasoning defaults

authored and
KRATOS
committed
Slice 4: sandbox depth — interactive/destructive command safety + confirmation policy

authored and
KRATOS
committed
Slice 5: agent guardrails — no-output guard + plan-enforcement reminders

authored and
KRATOS
committed
Slice 6: ask_user tool — agent asks clarifying questions mid-run

authored and
KRATOS
committed
Slice 7: agent modes — --mode flag + /mode command presets

authored and
KRATOS
committed
Slice 8: compaction execution — LLM-driven history compaction for long sessions

authored and
KRATOS
committed
Harden ask_user + write_file arg parsing for weak models (minimax-m3)

authored and
KRATOS
committed
Fix TUI: stop capturing the mouse so click/select/copy/paste work natively

authored and
KRATOS
committed
Add shared model-agnostic tool-arg tolerance layer (aliases) for all models

authored and
KRATOS
committed
Add repeated-tool-failure guard so no model loops on a bad tool call

authored and
KRATOS
committed
Zenline home/boot: uniform full-bleed themed background (no terminal-bg card)

authored and
KRATOS
committed
Coerce bool/int tool args from string/number forms (weak-model tolerance)

authored and
KRATOS
committed
Zenline rendering tier: markdown + syntax-highlighted code + colored diffs (PRD F4)

authored and
KRATOS
committed
Batch 2: slash-command autocomplete + interactive pickers (TUI)

authored and
KRATOS
committed
Batch 3: subagents — task tool spawns isolated sub-agent runs (PRD F11)

authored and
KRATOS
committed
Batch 4: plugin manifest enrichment + skills system (PRD F9/F15)

authored and
KRATOS
committed
audit fixes: sandbox + sessions (H2,H3,H4,M5,M6,M7 + lows)

authored and
KRATOS
committed
audit fixes: mcp + providers + zeroruntime (H6,H9,M4 + lows)

authored and
KRATOS
committed
audit fixes: zenline render (M8,M9,M10)

authored and
KRATOS
committed
audit fixes: tools (H1,M3 + lows)

authored and
KRATOS
committed
audit fixes: agent loop (M1,M2 + lows)

authored and
KRATOS
committed
audit fixes: tui + cli (H5,H7,H8,M11 + lows)

authored and
KRATOS
committed
audit: document plugin/skills extension wiring as discovery-only (deferred lows)

authored and
KRATOS
committed

Commits on Jun 7, 2026

address PR #103 review (CodeRabbit): OnUsage, multi-cancel flush, unsafe-toggle guard, hitbox clamp

authored and
KRATOS
committed
address PR #101 review (CodeRabbit): registry validation + test error checks

authored and
KRATOS
committed
Merge branch 'zenline-runtime-work' into audit-fixes

KRATOS
authored and
KRATOS
committed
Remove accidentally-committed shop.html test artifact + gitignore it

authored and
KRATOS
committed
Remove accidentally-committed audit doc from the repo (lives in issue #102)

authored and
KRATOS
committed
Remove brainstorming process specs from the repo (keep local)

authored and
KRATOS
committed
audit2: sandbox + sessions fixes (2 high, 2 med, 6 low)

authored and
KRATOS
committed
audit2: providers + runtime + zenline fixes (3 med, 7 low)

authored and
KRATOS
committed
audit2: tools fixes (4 high, 1 med)

authored and
KRATOS
committed
audit2: agent-loop fixes (1 med, 1 low)

authored and
KRATOS
committed
audit2: tui + mcp + skills fixes (2 med, 3 low)

authored and
KRATOS
committed
audit2: cli + modelregistry + dead-code cleanup (2 med, rest low)

authored and
KRATOS
committed