← ClaudeAtlas

aming-claw-hn-demo-during-worklisted

HN demo case for the fear that AI and subagents are black boxes during implementation. Guides evidence collection for timeline DAG, observer and subagent lanes, dispatch gate, evidence inspector, isolated worktrees, and append-only contract evidence.
amingclawdev/aming-claw · ★ 13 · AI & Automation · score 67
Install: claude install-skill amingclawdev/aming-claw
## REQUIRED FIRST READ Before any response that uses this skill, in this exact order: ListMcpResourcesTool() ReadMcpResourceTool(uri="aming-claw://current-context") ReadMcpResourceTool(uri="aming-claw://skill") ReadMcpResourceTool(uri="aming-claw://graph-first") current-context anchors project_id, governance URLs, and 3 guardrails. skill is the operating contract (Start Sequence, Observer Operating Modes). graph-first has copy-pasteable graph_query payload examples. Common failures when these are skipped: - Bootstrapping the wrong project (workspace_match auto-detected aming-claw) - Calling task_create dev/pm (V1 default is observer-led mf_parallel.v1) - Using Grep on the aming-claw codebase instead of graph_query - Fabricating trace_id strings (audit ledger is server-resolvable, will fail) - Running Execution Supervisor mode by default (Design Alignment is default) # HN Demo: During Work Show how Aming Claw makes subagent work observable while it is happening. ## Fear AI and subagents are black boxes: they run somewhere, change something, and the operator cannot tell whether scope, identity, or evidence is trustworthy. ## Evidence To Collect - Timeline DAG: ordered events for dispatch, implementation, verification, and handoff or review-ready state. - Observer/subagent lanes: which actor made each decision or produced each artifact. - Dispatch gate: evidence that branch, worktree, base commit, target commit, merge queue id, fence token, and owned file