Runs the human problem-discovery ceremony: grills problem, who has it, current alternatives, value proposition, and constraints, then hands a sharpened problem narrative to shape or authors project docs/BRIEF.md. Use when the user invokes pitch, starts work on a raw concept, or triage dispositions a spark or idea as pitch. Produces a problem-space narrative and a shape-now or park offer — never a bounded issue, criteria, or PRs. Not for quick capture (use idea), solution bounding (use shape), queue processing (use triage), or open-ended divergent inquiry (use explore as an agent technique when pitch reveals the direction is undecided).
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Runs the human problem-discovery ceremony: grills problem, who has it, current alternatives, value proposition, and constraints, then hands a sharpened problem narrative to shape or authors project docs/BRIEF.md. Use when the user invokes pitch, starts work on a raw concept, or triage dispositions a spark or idea as pitch. Produces a problem-space narrative and a shape-now or park offer — never a bounded issue, criteria, or PRs. Not for quick capture (use idea), solution bounding (use shape), queue processing (use triage), or open-ended divergent inquiry (use explore as an agent technique when pitch reveals the direction is undecided).
Human problem-discovery ceremony. Narrows sparks and ideas into a framed problem so shape can mint an issue, and bootstrap can consume a pitched project BRIEF.
Contents
Critical Rules
Verification
Quick Reference
Process
Related Skills
Topics
Input: $ARGUMENTS
Critical Rules
Agents never initiate a pitch. This ceremony is human-invoked only. On Claude Code the sidecar sets disable-model-invocation: true; on every target this rule binds behaviorally. Agent legwork inside a human-opened pitch (competitive scans, file writes the skill directs) is fine — opening one is not.
Log invocation first — loaf journal log "skill(pitch): <idea, problem, spark, or intake item>" before interviewing.
Problem-space only — grill what, who, and why-valuable. Approach, architecture, decomposition, and verification design belong to shape. A narrative that reads like a pseudo-shape is a failure; rewrite before landing.
One question at a time, recommendation-first — using your harness's structured question tool if it has one (otherwise one inline question per message). Never a multi-field form. Order by impact on the narrative. Full mechanics: references/interview-guide.md.
Never bound, never ship — do not add definition-of-done criteria, do not write an out-of-scope statement, do not run loaf issue check or loaf issue promote, do not push, do not open PRs. Never auto-run shape or bootstrap.
Shape mints on the happy path — same-session shape-now hands the authored narrative; shape runs loaf issue new with that body. Pitch writes an issue body only when parking an unshaped row or when $ARGUMENTS already names an issue (loaf issue edit replaces the body).
Titles name the concept — propose a working title, never another work unit's alias. Provenance lives in the issue row, the spark/idea resolution, and frontmatter on docs/BRIEF.md.
Log the outcome — loaf journal log "decision(pitch): <title, ref, or project> — <shape-now|park-issue|park-idea|handed-to-bootstrap>".
Verification
Issue-scale: a problem narrative exists against the shared skeleton; it was handed to shape, written into an existing issue body, or minted as an unshaped triage row with that body and no criteria
Project-scale: docs/BRIEF.md exists with source: pitch and the shared problem-space skeleton
Cold-read: problem, who, alternative, and value nameable in one pass; zero solution-space content; no out-of-scope statement and no criteria added by this skill
Named sparks were promoted to an idea when pitching them; ideas and sparks were resolved against the issue only after a row exists
No push; no PR; shape and bootstrap were not auto-run
Journal shows skill invocation and outcome entries
Quick Reference
Harness
Invoke skill
Claude Code (plugin)
/loaf:pitch
OpenCode, Cursor, Codex, Amp
/pitch
Scale detection
Signal
Scale
Output
Existing project (git history, source, or Loaf state) + a concept
Issue
Problem narrative → shape (loaf issue new --body) or an unshaped triage row
Empty or minimal directory / greenfield product pitch
Project
docs/BRIEF.md with source: pitch → hand to bootstrap
Detect and confirm briefly; let the human correct. When both could apply (repo exists but they want a new product pitch), ask once.
Landing offers
Offer
When to recommend
What pitch does
Shape now
Framing is solid; they want to bound next
Hand the narrative; do not mint; do not auto-run shape
Park as issue
Framed, durable, not bounding yet
loaf issue new "<title>" --body - with the narrative only; status stays triage
Park as idea
Too thin to keep as a row, or might discard
loaf idea capture --title "..."; journal the gist
Hand to bootstrap
Project-scale BRIEF authored
Point at bootstrap; do not auto-run it
Pitch never pushes; never opens PRs. There is nothing to commit at issue scale — the row lives in SQLite. Project-scale may commit docs/BRIEF.md if the human wants it durable.
Spark and idea promotion
Input
Read
Then
Spark
loaf spark show <ref>
loaf idea capture --title "..." then loaf spark promote <spark> --to-idea <idea>; grill from the idea
Idea
loaf idea show <ref>
Grill; after a row exists, loaf idea resolve <idea> --by <ref>
Existing issue
loaf issue show <ref>
Grill; loaf issue edit <ref> --body - writes the narrative (replaces the whole body)
Free text
—
Grill; shape-now hands text; park captures an idea or mints an unshaped row
Do not invent a pitch from the queue without human selection. When they name an intake item, read it (loaf intake list / the item's read command).
loaf idea promote --to-spec is not this path. Resolve ideas against the minted issue.
Problem-narrative skeleton
Author against these sections, problem-space sentences only. This text is what shape puts in --body (or what a park-as-issue row stores):
## Problem Statement## Who Has It## Current Alternatives## Value Proposition## Constraints## Sequencing and Relationships## Sources and Research Links## Open Questions
Do not add an out-of-scope statement. Shape bounds; pitch frames.
Defined terms
Problem narrative — pitch's issue-scale output. Superseded as the working surface once shape mints and bounds the issue; may accrete parked problem-space sentences until then.
BRIEF — project-scale docs/BRIEF.md. A project document, not a work container.
Accretion — adding problem-space concepts to a parked narrative is legal; solution prose is not.
Shape now — hand the narrative to shape, which mints via loaf issue new and owns bounding.
Process
Step 1: Log and parse input
loaf journal log"skill(pitch): <idea, problem, spark, or intake item>"
Parse $ARGUMENTS: free text, a spark, an idea, an issue ref, an intake ref the human already chose, or empty (ask what to pitch). Read the named item when provided. Do not invent a pitch from the queue without human selection.
Step 2: Detect scale
Apply the Quick Reference table. Confirm: "I'll treat this as an issue-scale pitch on this repo" or "…as a project-scale pitch for a new BRIEF." Adjust if corrected.
Step 3: Promote sparks; read ideas
When the named input is a spark, promote it to an idea before grilling so the capture trail is one idea, not a dangling spark:
Pin a one-or-two-line destination before dimension grilling (fixes narrative scope; project scale feeds VISION success criteria; issue scale sharpens what good looks like for the row)
Dimensions: problem, who has it, current alternatives / competitive landscape, value proposition, constraints (plus sequencing and open questions when needed)
Depth: scenario stress-testing, challenge stance, glossary-term hygiene; open questions must pass the specifiability test and carry HITL/AFK tags when precise
Applicability judgment: skip formal competitive analysis and deep personas when the pitch kind does not warrant them (bug fixes, internal chores)
One question at a time, recommendation-first, ordered by narrative impact
Stop on exit criteria or when answers stop changing the framing
If the direction is genuinely undecided mid-interview, offer the explore technique (agent-side; Explorations and checkpoints) rather than forcing a false narrative. Pitch remains the human front door; explore is not re-opened as a slash alternative by this skill.
Step 5: Evidence delegation (when warranted)
When competitive or landscape facts would change the narrative and are not already known:
Spawn a researcher subagent with a bounded question (competitors, substitutes, prior art — not solution design).
Land evidence:
Issue scale: source links in the narrative's Sources and Research Links. If a longer scan lands on disk, name it for the landscape, never for the work unit, and cite it from Sources.
Project scale: inline source links in docs/BRIEF.md Sources and Research Links.
Resume the interview or draft with recommendations informed by the scan.
Never fabricate competitive claims. If a scan is skipped, say so in Sources ("no external scan; alternative is internal workaround X").
Step 6a: Issue-scale ceremony
Propose a working title — names the concept locally. Confirm with the human. This becomes shape's loaf issue new title (or the park-as-issue title).
Author the problem narrative against the skeleton above. Problem-space sentences only.
Accretion note — tell the human: parked problem-space concepts may accrete until shaping starts; once the issue is minted, the body is the home.
Cold-read the narrative (interview guide test); revise with the human until it passes.
Offer landing (recommendation-first) using the Landing offers table.
Execute the chosen landing:
Shape now: hand the full narrative and any spark/idea refs. Shape runs loaf issue new "<title>" --body - (or --body-file) with that text. Do not mint, do not add criteria, do not open a PR. After shape mints, resolve intake: loaf idea resolve <idea> --by <ref> (and loaf spark resolve <spark> --by <ref> only if the spark was never promoted).
Park as issue: mint the unshaped row yourself, then resolve intake against it:
loaf issue new "<title>" --body -
loaf idea resolve <idea> --by <ref>
Paste the narrative on stdin. Do not add criteria. Do not write out-of-scope. Default status is triage. Read back with loaf issue show <ref>.
Park as idea: if no idea exists yet, loaf idea capture --title "<title>". Journal the gist (loaf journal log "discover(pitch): <one-line problem>"). Do not mint an issue.
Existing issue:loaf issue edit <ref> --body - with the full narrative. Edit replaces the body; do not strip a row that is already bounded — if criteria already exist, hand the narrative to the human and let shape merge.
Closing ceremony (required — never trail off). After the landing is executed, announce completion with a full closing block:
Recap the narrative — section-by-section gist (Problem, Who, Alternatives, Value, Constraints, Sequencing, Open Questions). Name where it lives (handed to shape, unshaped issue <ref>, idea <ref>, or the conversation plus journal gist).
Restate the landing actually taken and what it means next:
Shape now — run shape next to mint the issue from this narrative and bound implementation. No row was minted here.
Park as issue — <ref> holds the problem in its body and is unshaped; run shape later on that ref.
Park as idea — the idea remains open; re-invoke pitch or shape when ready. Name the idea ref.
Announce completion in plain language: "Pitch is complete." Do not end on a dangling offer or an unfinished sentence.
Step 6b: Project-scale ceremony
Author docs/BRIEF.md using bootstrap's brief skeleton with frontmatter:
---source:pitchcreated:<ISO-8601>archived:true---
Same problem-space sections as issue scale, at project altitude (Sequencing describes the initial arc as prose).
Cold-read and revise with the human.
Optionally commit docs/BRIEF.md if the human wants it durable before bootstrap; still no push unless they ask outside this skill's duties — pitch itself never pushes.
Closing ceremony (required — never trail off). Announce completion with a full closing block — do not hand off in a half-sentence:
Recap what was authored — section-by-section gist of the BRIEF (Problem Statement, Who Has It, Current Alternatives, Value Proposition, Constraints, Sequencing and Relationships, Sources and Research Links, Open Questions). One or two sentences per section is enough; the human should hear what landed without reopening the file.
Artifact path — name docs/BRIEF.md explicitly, including that frontmatter carries source: pitch.
Explicit handoff — state, verbatim in spirit: next, run bootstrap: it reads this BRIEF as discovery-done (source: pitch), interviews only on gaps, and populates the operating documents (VISION, STRATEGY, ARCHITECTURE, AGENTS). Do not auto-run bootstrap.
Announce completion in plain language: "Pitch is complete." The ceremony ends with a period, never a trail-off.
Step 7: Log the outcome
loaf journal log"decision(pitch): <title, ref, or project> — <shape-now|park-issue|park-idea|handed-to-bootstrap>"
The journal line is mechanical; the human-facing close is the closing ceremony in Step 6a/6b. Never log-and-stop without that recap and next-step restatement.
Related Skills
shape — solution-space bounding; mints the issue from the problem narrative (loaf issue new) and owns criteria, out-of-scope, and decomposition
bootstrap — consumes docs/BRIEF.md (source: pitch → gap interview) and populates operating documents
triage — queue dispositions; may hand a spark or idea to pitch when problem discovery is needed
explore — agent-side technique when pitch finds the direction still undecided
idea — quick capture without ceremony; not a substitute for pitch
research — patterns the researcher subagent follows for landscape scans
Name every on-disk artifact for what it is, never for the work unit that produced it. The issue row or docs/BRIEF.md already records provenance. Put source in frontmatter, not the filename. See the foundations skill; loaf check --hook artifact-names enforces it at commit.
