From hw
Resume Hypo-Workflow execution from the saved state when the user wants to continue an interrupted or stopped pipeline.
npx claudepluginhub hypoxanthineovo/hypo-workflow --plugin hwThis skill uses the workspace's default tool permissions.
📌 输出语言规则:
Start Hypo-Workflow execution when the user wants to begin running milestones, continue automatically through the pipeline, or execute the first prompt.
Resumes interrupted HOTL workflow runs: loads .hotl/state JSON state, verifies last step before continuing. Resolves by run ID or workflow path, detects stale runs, lists multiples.
Resumes interrupted analysis pipelines by reading pipeline state from working directories, migrating V1 to V2 formats if needed, and continuing from next ready agents using DAG walker. Useful after context limits, failures, or breaks.
Share bugs, ideas, or general feedback.
📌 输出语言规则: 读取 config.yaml → output.language
Use this skill to continue from .pipeline/state.yaml without restarting completed work.
.pipeline/state.yaml existspipeline.status=running or pipeline.status=stopped~/.hypo-workflow/config.yaml if present..pipeline/config.yaml and .pipeline/state.yaml.current.phase=needs_revision, resume the revision path using acceptance.feedback_ref as input; do not continue from a previously completed step.current.phase=follow_up_planning, start the active continuation or the matching cycle.continuations[] follow-up plan.current.prompt_file, current.step, and current.step_index still point to a valid prompt and step for ordinary execution phases..pipeline/.lock exists, parse it as an execution lease.
lease_takeover evidence, and record inferred_stallreported_failure: take over and preserve the failure evidencecurrent.phase=executing and update top-level last_heartbeat./hypo-workflow:start:.pipeline/PROGRESS.md, .pipeline/log.yaml, .pipeline/state.yaml, and last_heartbeat after each meaningful transition..pipeline/derived-refresh.yaml, and surface repair guidance instead of rolling back the lifecycle write./hw:compact automatically when compact.auto=true; skip it when compact.auto=false..pipeline/feature-queue.yaml exists, resume batch auto-chain from the saved state:gate: confirm by pausing before the next Featurejust_in_time, decompose it only after it becomes current.pipeline/metrics.yaml, preserving n/a for unavailable telemetryretry / deferred / stop decision model on failures..pipeline/.lock when the resume turn completes, stops, blocks, aborts, or finishes..pipeline/.lock lease says another execution is activeWatchdog-triggered resumes follow the same safety rules as user-triggered resumes. They must update last_heartbeat, honor fresh execution leases, take over expired leases with evidence, and write a concise note to .pipeline/watchdog.log when the triggering script provides context.
references/state-contract.md — resume semantics and required fieldsreferences/commands-spec.md — command behaviorreferences/progress-spec.md — progress summary rulesreferences/config-spec.md — global/project config fallback rulesSKILL.md — full execution context if needed