docs: space autonomy gap analysis (MiniMax-M2-7-HighSpeed)

[lsm]

Summary

Comprehensive gap analysis of the space/task/workflow system covering semi-autonomous and human-in-the-loop behaviors. Identifies 20 open gaps total — 10 new findings + 10 previously tracked items.

New critical gaps

• Gap A: request_human_input tasks have no unblock path — no RPC to deliver a human's answer and resume the agent
• Gap E: No spaceTask.sendHumanMessage equivalent (Room has task.sendHumanMessage at task-handlers.ts:1030, Space has no equivalent)
• Gap B / chore: bump version to 0.1.1 #18: No task retry RPC — only cancel+new-run which discards all workflow state
• Gap H / fix: make publish steps idempotent for re-runs #10: Static workflow topology — no conditional branching by autonomy level

New medium/low gaps

• Gap C: Completion action failure doesn't set blockReason: 'execution_failed'
• Gap D: Completion action approval has no comment field (only rejection does)
• Gap F/fix: use npm Trusted Publishing instead of NPM_TOKEN #6: No autonomy-differentiated retry policy
• Gap G/fix: make platform package publish resilient to partial releases #9: No review SLA/timeout
• Gap I/feat: split E2E tests into no-LLM and LLM groups for better parallelism #16: Multi-gate blockingGateId missing
• Gap J/feat: init git submodules in worktrees and add loading spinners for archive/delete #17: No consecutive failure escalation across runs

Tracking doc

docs/space-autonomy-gap-analysis-minimax-m2-7-highspeed.md

(cross-references docs/space-autonomy-hitl-gaps-claude-opus-4-6.md for the 19-gap prior analysis)