Inno Pipeline Planner
Productivityinno-pipeline-plannerInno Pipeline Planner is an Agent Skill for productivity published in the OpenLAIR/dr-claw repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Guides the user through an interactive conversation to define their research project, then generates research_brief.json and tasks.json. Use when starting a new project, when no research_brief.json exists, when the user wants to start from a specific pipeline stage, or when the user wants to redefine their research pipeline.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/OpenLAIR/dr-claw/blob/HEAD/skills/inno-pipeline-planner/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/inno-pipeline-planner, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Inno Pipeline Planner
Run an interactive planning flow that turns user conversation into:
.pipeline/docs/research_brief.json.pipeline/tasks/tasks.json
Keep this file short. Load full schemas and field-level rules from:
references/pipeline-contract.md(index)
Read only what you need:
references/generation-rules.md: generation logic, ordering, dependencies,nextActionPromptreferences/brief-schema.md:.pipeline/docs/research_brief.jsoncontractreferences/tasks-schema.md:.pipeline/tasks/tasks.jsoncontract
Non-negotiables
- Work only inside the current project directory.
- Do not fabricate papers, datasets, metrics, or results.
- Ask follow-up questions when information is vague; do not guess.
- Ask in small batches (2-3 questions), not a long static form.
Workflow
1) Inspect existing pipeline state
Check:
.pipeline/docs/research_brief.json.pipeline/tasks/tasks.jsoninstance.json(legacy source)- Content in
Survey/,Ideation/,Experiment/,Publication/, andPromotion/directories (to detect pre-existing artifacts)
If brief exists, summarize title, goal, current startStage, and completion status, then ask:
- Refine existing brief/tasks
- Regenerate from scratch
- Change the starting stage
2) Collect project context via conversation
Capture at least:
- Topic/problem
- Goal or hypothesis
- Success criteria or evaluation signal
- Current survey depth or known reference set
Determine the starting stage early in the conversation:
- Ask what the user already has: "Do you already have a research idea, experimental results, or are you starting from scratch?"
- If the user mainly needs literature review, gap analysis, or reference collection ->
startStage = "survey" - If the user has a concrete idea with problem framing and success criteria ->
startStage = "experiment" - If the user has experimental results and analysis ->
startStage = "publication" - If the user already has a paper/manuscript and mainly needs a homepage, slide deck, narration, or demo assets ->
startStage = "promotion" - If the user is starting from scratch or only has a vague direction ->
startStage = "survey"(default) - Detect automatically from conversation context (e.g., "I already ran all experiments" implies publication; "I need slides for my paper" implies promotion).
Typical question buckets:
- Project identity: topic, prior paper/method/dataset, target venue (optional)
- Scope and method: core question, approach, expected outcome
- Evaluation: data source, metrics/protocol, baseline expectations
Adapt to context:
- Skip already-provided details.
- Skip questions for stages before
startStage: If starting from experiment, do not ask survey or ideation questions in detail — just capture a brief summary of the existing context in those sections. - If exploratory, keep experiment/publication/promotion sections lightweight.
- If user provides concrete plan, prepare for
pipeline.mode = "plan"; otherwise use"idea".
3) Write pipeline files
Create if missing:
.pipeline/config.json.pipeline/docs/research_brief.json.pipeline/tasks/tasks.json
Use the exact JSON contracts and generation rules in:
references/pipeline-contract.mdand linked reference files
Rules:
- Set
pipeline.startStageto the determined starting stage (default:"survey"). - Generate tasks only for stages >=
startStagein the stage order (survey < ideation < experiment < publication < promotion). - For skipped stages: still populate their
sections.*fields in the brief with whatever context the user provided, but do not create task blueprints or tasks for them. - Tailor blueprint titles/descriptions to the user topic (never generic filler).
- Keep quality gates domain-appropriate.
- Resolve recommended skills from local available skills (
.agents/skills/orskills/), optionally usingstage-skill-map.jsonif present.
4) Summarize and confirm next action
After writing files, present:
- Brief summary (title, goal, starting stage, filled vs missing sections)
- Task overview (count by stage + first 2-3 task titles per stage) — only for active stages
- Recommended first task and why
5) Handle iteration requests
If user asks for updates:
- Update brief content directly when only text/content changes.
- Regenerate
tasks.jsonwhen pipeline structure/blueprints/stages change. - If user asks to change the starting stage: update
pipeline.startStagein the brief, then regeneratetasks.jsonto include only the active stages. - If asked to add one task only, append a single task with next numeric
idinstead of full regeneration.