ideation-funnel

Ideation Funnel (orchestrator)

Runs a blog-post ideation pipeline over one or more braindump notes. This skill is the orchestration glue; the actual procedure lives in ideation-funnel.md in this skill's directory and is the single source of truth — read and follow it, do not reinvent or paraphrase the stages.

Setup: locate the notes directory

This skill operates on a folder of markdown notes (e.g. an Obsidian vault, or any directory of .md files). Determine it before running:

• If the user named a directory, or it is clear from the conversation, use that.
• Otherwise ask once: "Which notes folder should I work in?"

Inside that folder the skill maintains files it creates on first use: a backlog file Post backlog.md (the seed queue) and one <Topic> — funnel.md working file per run. Raw braindump notes are read-only.

Bundled docs (read from this skill's own directory)

• ideation-funnel.md — the five-stage procedure plus the Stage 3.5 red-team. Follow it exactly.
• drafting-checklist.md — craft reference for the Stage 4 depth gate and for the drafting handoff.

Inputs

One or more braindump note titles or paths. If several are given, atomize across all of them into a single clustering pass — cross-note clusters are often the best output. If none are named, ask which note(s) to run.

Orchestration

1. Load context. Read ideation-funnel.md and drafting-checklist.md from this skill's directory, every named braindump, and the notes folder's Post backlog.md if it exists.
2. Related-file discovery (propose, never auto-ingest). Scan the notes folder for notes the braindump(s) wikilink to, or that share key terms. Present the candidates and ask the user which to pull in as extra context. Pull in only what they pick.
3. Run Stages 1 through 3.5 from ideation-funnel.md across all inputs in one pass: atomize (C/Q/R/L tags), cluster with the cohesion/coupling/bridge/right-size checks, thesis-test each cluster, then red-team each survivor. Candidates failing falsifiability or taboo drop to Research targets.
4. Run Stage 4 up to the reserved calls, then STOP. Fill the depth-gate worksheet for each live candidate and propose Field scores. Do not set Strategic scores. Do not pick the winner. Present the score table plus the depth-gate worksheets and ask the user for (a) the Strategic scores and (b) the winning cluster.
5. After the user answers, write the Stage 5 scope brief for the chosen winner, ordering beats by what the depth gate found load-bearing.
6. Write artifacts. Create <Topic> — funnel.md as a sibling of the raw note. Update Post backlog.md with parked / queued / research-target items, deduping against what is already there.
7. Hand off. Point the user to drafting-checklist.md for drafting. Do not start a draft unless explicitly asked.

Hard rules

• Never edit a raw braindump note. It is canonical. All output goes to the <Topic> — funnel.md working file or the backlog.
• Never set Strategic scores or pick the winner yourself. Those are the user's calls. Stop and ask (step 4).
• Never invent citations or links. Preserve unresolved references and placeholder URLs verbatim and flag them.
• Always dedupe before appending to the backlog.

Red flags — stop and correct

• About to write a strategic score or declare a winner without the user → STOP, ask.
• About to edit the raw note → STOP, write to the working file instead.
• About to silently read a pile of "related" notes → STOP, propose them first.
• About to fabricate a source for a placeholder link → STOP, flag it instead.