process-mining-assistant
通过命令行工作流程执行端到端的过程挖掘分析,逐步摄取、分析、清理、挖掘和报告事件日志,使用PM4Py。工作流程生成基于阶段的工件(包括版本化笔记本),并在决策检查点暂停,以便用户验证发现并选择如何继续。
name: pm-00-orchestrator description: Orchestrate an end-to-end process mining engagement by chaining phase skills with decision checkpoints and reproducibility safeguards. metadata: short-description: End-to-end orchestrator for the process mining skill suite.
Overview
This skill coordinates the full process mining workflow from environment validation to reporting.
It enforces decision checkpoints, stage gating, and reproducibility as defined in pm-99-utils-and-standards.
Use it when the user wants an end-to-end run or a guided resume from a specific stage.
When to use this skill
Use when the user requests an end-to-end process mining run or wants help sequencing phases. Examples:
- “Run the full pipeline for this log.”
- “Resume from performance analysis.”
Inputs required
- Input log file path and format
- Optional config file (YAML or JSON)
- Output directory
- Stage resume point if resuming
Outputs produced
- Stage artefacts in
output/stage_*/ - Versioned notebooks in
output/notebooks/Rx.xx/ - Updated
output/manifest.json - Final report artefacts in
output/stage_09_report/
Workflow
- Confirm objective, scope, and success criteria.
- Run environment detection and validation via
pm-01-env. - Run phases sequentially by delegating to
pm-02throughpm-10. - Enforce decision checkpoints and apply phase gating.
- On each transition, check notebook and artefact hashes and update the manifest.
- If strict reproducibility is enabled, stop on any mismatch and request action.
Decision checkpoints
Ask: Choose the run mode for this engagement.
Complication: The run mode affects reproducibility and how we handle changes between stages.
Options:
- Strict reproducibility with hash enforcement [preferred]
- Standard reproducibility with warnings
- Ad-hoc exploratory run with no hash enforcement
Impact:
- Option 1: stop on any mismatch, enforce revision bumps and re-runs.
- Option 2: allow continuation with logged warnings and revision bumps.
- Option 3: fastest iteration but lowest auditability.
Ask: Choose the execution approach for the pipeline.
Complication: Using stage-by-stage execution gives tighter control, while full pipeline runs are faster but less flexible.
Options:
- Stage-by-stage with checkpoints [preferred]
- Full pipeline run via
run_pipeline.py
Impact:
- Option 1: explicit decisions between stages, higher governance.
- Option 2: faster execution, fewer intervention points.
Commands
- End-to-end pipeline:
python .codex/skills/pm-00-orchestrator/scripts/run_pipeline.py --file <path> --format <csv|xes> --output <dir>
- Resume from stage (manual sequence):
python .codex/skills/pm-01-env/scripts/00_detect_env.py --output <dir>python .codex/skills/pm-01-env/scripts/00_validate_env.py --output <dir> --setup-venv --venv-dir .venv --requirements .codex/skills/pm-99-utils-and-standards/requirements.txt- Follow phase skills in order with their command patterns.
Validations
- Confirm
manifest.jsonexists or will be created in the output root. - Confirm stage outputs match required artefact folders.
- Exit criteria: all requested phases complete with artefacts and an updated manifest.
Issue handling
- On errors, check
references/index.jsonfor matching error signatures and follow the referenced runbook. - If
references/index.jsonorreferences/README.mdis missing, create them using the pm-01-env issue pack format. - When you resolve a new issue, add a runbook under
references/issue-fixes/and register it inreferences/index.json.
Failure modes and remediation
- Missing dependencies: stop and route to
pm-01-envremediation. - Hash mismatch in strict mode: stop, bump revision, re-run impacted stages.
- Missing stage artefacts: re-run the missing stage via its skill.
Compatibility notes
- Delegates to
pm-01throughpm-10and followspm-99standards. - Compatible with
process_mining_cli.pywrapper if present.
Version history
- R1.00 Initial orchestrator split from the monolithic process-mining-assistant.