v0.2.0 — Permission-first runtime for Lark-native agents

What's new in v0.2.0

LARC is the permission-first runtime layer between OpenClaw/Claude Code agents and Lark (Feishu). This release completes the core queue lifecycle, auth router, and Windows support milestone.

Highlights

• larc auth suggest — infers minimum required scopes + identity type + gate level from any natural-language task description (32 task types)
• larc auth router — per-operation routing: user_token | bot_token | blocked (some Lark ops require User OAuth and cannot be automated — these are intentionally surfaced, not silently failed)
• Full queue lifecycle — enqueue → in_progress → done/failed/blocked with prune, retry, stats, recover, delegate, context, openclaw subcommands
• larc ingress openclaw — returns a next-action bundle in OpenClaw format so agents can pick up and execute tasks without polling
• Windows support — cross-platform shims (realpath, stat, sed, date), PowerShell launcher (bin/larc.ps1), fcntl fallback, CI matrix on windows-latest
• larc quickstart — 7-step idempotent onboarding: token check → Base provisioning → Drive workspace → agent registration → queue smoke test
• GitNexus skills — 6 Claude Code skills for code intelligence (.claude/skills/gitnexus/)

Key files

File	Purpose
config/scope-map.json	32 task types × required scopes × 4 profiles
config/gate-policy.json	32 task types × risk × gate (none/preview/approval)
lib/ingress.sh	Queue engine — full lifecycle + OpenClaw bridge
lib/auth.sh	Scope inference + token routing
PLAYBOOK.md	Operational runbook — setup through business launch

What's next (v0.3.0)

• Stabilize IM daemon loop (PID management + auto-restart) — T-015
• MergeGate integration (lib/mergegate.sh) — T-018
• OpenClaw-assisted autonomous mode environment verification — T-020
• Expand scope-map from 32 → 50 task types — T-017

---

English · 简体中文 README · 日本語 README

v0.1.1

Summary

This release aligns the runtime with verified Lark API usage after v0.1.0.

Included

• Fixed memory sync to parse the current Lark Base table response shape
• Fixed send flow to use the correct Lark IM shortcut command
• Improved YAML-based batch agent registration flow
• Cleaned up naming hygiene and synced README/PLAYBOOK state after the initial public release

Verification

• register-agents dry-run generated correct commands for all configured agents
• larc memory pull --agent main succeeded against Lark Base
• larc send succeeded against Lark IM and returned a run ID

Notes

Internal planning artifacts remain untracked and are not part of this release.

v0.1.0

LARC is opening publicly with an initial docs-first slice and a first runtime core freeze.

LARC explores a permission-first runtime for Lark-native office-work agents, treating Lark Drive, Base, IM, Approval, and Wiki as operating surfaces rather than simple external integrations.

This release includes:

• trilingual README entry points
• trilingual CONTRIBUTING guides
• trilingual terminology glossaries
• permission model and auth-suggest cases
• release-readiness and launch messaging docs
• an initial runtime core around permission inference, approval gates, and agent registry behavior

This release should not be read as a fully frozen end-state runtime. It is the first public milestone of the project, with later phases still continuing around memory, send flows, agent registration, and broader runtime packaging.