productfoo 0.0.0-stage → 1.0.1

package/skills/generate-gtm-brief/references/gtm-brief-template.md

@@ -0,0 +1,56 @@
+# Go-to-Market Brief
+
+## Decision
+
+- Launch decision:
+- Decision owner:
+- Review state:
+- Target window:
+
+## Product Truth
+
+- What changed:
+- What did not change:
+- What works now:
+- Known limits:
+
+## Audience and Pain
+
+- Primary audience:
+- Pain or unmet need:
+- Current alternative:
+- Evidence for the need:
+
+## Positioning and Message
+
+- Positioning statement:
+- Differentiated value:
+- Proof points:
+- Claims to avoid:
+
+## Rollout and Readiness
+
+- Release stages:
+- Entry and exit criteria:
+- Required product, support, and communication artifacts:
+- Owners and approval boundaries:
+
+## Measures
+
+- Desired outcome:
+- Leading signal:
+- Success measure:
+- Stop or rollback signal:
+
+## Evidence Ledger
+
+- **Sources used** — approved evidence that supports this brief.
+- **Sources omitted** — relevant evidence lanes not consulted.
+- **Trust status** — fact, approved evidence, inference, or hypothesis.
+- **Limits** — unresolved product, audience, or launch uncertainty.
+
+## Human Decisions
+
+- Decisions already made:
+- Decisions still required:
+- Recommended next decision:

package/skills/gong-quotes/SKILL.md

@@ -0,0 +1,43 @@
+---
+name: gong-quotes
+description: "Use when approved customer-call evidence needs traceable quotations, paraphrases, themes, or quote-gap analysis."
+---
+
+# Gong Quotes
+
+## Usage
+
+The name is retained for compatibility; the workflow is provider-neutral. Work
+only from approved sanitized evidence. Do not open, paste, commit, or reproduce a
+raw transcript.
+
+1. Confirm the topic, study or packet scope, and whether exact quotations are required.
+2. Use `foo discovery` to match and apply an approved call through the in-memory
+   proof gate when raw call access is needed. Consume only the resulting approved
+   sanitized evidence envelope.
+3. Reproduce a verbatim quote only when those exact words are already present in
+   approved sanitized evidence. Otherwise write a labeled paraphrase; never
+   reconstruct or invent wording.
+4. Group quotes by theme, retain the stable evidence reference and date, show
+   dissent, and avoid participant PII.
+5. Return the quote set in chat or an explicitly requested local artifact; never
+   update a CRM, call system, or message destination.
+
+### Source ledger
+
+- **Sources used** — approved sanitized evidence references and quote eligibility.
+- **Sources omitted** — calls or evidence lanes not approved, unavailable, or out of scope.
+- **Trust status** — exact approved quote, labeled paraphrase, or unsupported gap.
+- **Limits** — coverage, representativeness, and wording that could not be quoted.
+
+Optional omissions are named in both the human summary and the artifact.
+
+## Error Handling
+
+- If a required call lacks source approval, provider access, or an approved
+  sanitized envelope, stop with `status: blocked`, name what did not run, give
+  the exact `next action` from `foo discovery`, and set `exit: nonzero`.
+- If an optional call is omitted, continue only when the remaining evidence can
+  support an honest result and list the omission.
+- If asked for a quote that does not exist in approved evidence, return a quote
+  gap. Do not use a generic quote or model-generated substitute.

package/skills/gong-quotes/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Gong Quotes"
+  short_description: "Extract traceable quotes only from approved sanitized evidence."
+  default_prompt: "Use $gong-quotes for the relevant workflow."

package/skills/grill-me/SKILL.md

@@ -0,0 +1,27 @@
+---
+name: grill-me
+description: Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
+---
+
+## Human-facing voice
+
+Before writing for a person, load `foo skill product-copywriting/references/foo-voice.md`
+and apply it within this skill's required output format. Keep the evidence record
+and workflow responsibilities intact. If the reference cannot be loaded, report
+the delivery gap and preserve evidence, uncertainty and dissent explicitly.
+
+## Usage
+
+Use this skill when its frontmatter description matches the user request. Follow the workflow and constraints below.
+
+
+When the user explicitly asks to be grilled, pressure-test the plan until its consequential decisions are understood, resolving dependencies one at a time. When another skill borrows this interaction model, use its scope and question budget: one question, a recommended answer, then wait. Borrowing the pattern does not request an exhaustive interview. Keep criticism specific to the plan and its evidence. For each question, provide your recommended answer.
+
+Ask the questions one at a time.
+
+If a question can be answered by exploring the codebase, explore the codebase instead.
+
+## Error Handling
+
+- If required files, tools, connectors, credentials, or source artifacts are unavailable, state the blocker clearly and continue only with safe local evidence.
+- If the user request requires missing product judgment or source-boundary approval, ask one focused question before proceeding.

package/skills/grill-me/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Grill Me"
+  short_description: "Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the"
+  default_prompt: "Use $grill-me for the relevant workflow."

package/skills/ideate/SKILL.md

@@ -0,0 +1,180 @@
+---
+name: ideate
+description: "Leaf craft for productfoo's ideation runs: how to think as an ideator, a stress reviewer, or a judge inside the run contract that `foo ideate` emits. The conductor (CLI) owns the choreography — phases, cells, blindness, budgets, verification; this skill holds only the per-role thinking moves."
+---
+
+# ideate (leaf craft)
+
+`foo ideate "<idea>"` emits the run contract: the phases, the per-ideator
+frame × lens × seed cells, the protocol rules, and the record scaffold.
+Execute that contract as printed. This skill adds only the craft — how to be
+good *inside* each role. It contains no orchestration; if this file and the
+emitted contract ever disagree about sequencing, the contract wins.
+
+## Ideator craft (DIVERGE)
+
+Work only your own cell; the contract tells you your frame, lens, and seed.
+Use the shared prior-work brief as context, including previous ideas that
+failed or were deferred. Compare with the strongest earlier proposal. A useful
+new direction changes a mechanism, resolves an old constraint, or combines
+findings into something the user can now do. Explain that difference with a
+concrete before/after example in `mechanism`; a renamed earlier idea is not new.
+Research conclusions retain their uncertainty until checked against evidence.
+The thinking moves, each with its why (every depth rotates all five):
+
+- **analogy** — steal the *mechanism* from another domain, never the surface;
+  surface-analogies decorate, mechanism-analogies transfer.
+- **inversion** — "how would we guarantee the failure?" then negate each
+  ingredient; failure is easier to enumerate honestly than success.
+- **how-might-we** — phrase the tension as HMW questions, then answer the
+  sharpest; the question shape stops premature solutioning.
+- **first-principles** — separate bedrock truths from inherited assumptions
+  and rebuild only from bedrock; most designs smuggle in the defaults.
+- **second-order** — "and then what?" twice; first-order wins often create
+  the next problem.
+
+Return every idea as an object — `claim, frame, lens, reasoning, mechanism,
+assumption, novelty, claims_relied_on` — appended to the record's ideas block.
+Idea and evaluation stay separate so raw claims survive scoring. `reasoning`
+is the frame's move, shown, in the tagged shape the contract prints (an
+inversion enumerates its failure ingredients before the negation; an analogy
+names its source domain and the mechanism stolen). Put a concise rationale
+in `reasoning`, not a private reasoning transcript. The check verifies the
+shape of the explanation; it cannot certify originality or usefulness.
+
+Return at most 4 new ideas. Retain the strongest existing direction as the
+comparison baseline named by the conductor. For new candidates, discard generic
+restatements of that baseline and return ideas whose `novelty` can honestly complete `vs: <nearest existing or obvious
+approach> — <what this beats it on>`. Constraints stated in the idea are held
+once at contract level; never restate consent, approval, or exclusion language
+inside your objects — a mechanism spent on disclaimers is a mechanism not
+spent on the idea.
+
+Write the claim as one sentence and put the texture in `mechanism`: it is free
+text, read by humans, and never clustered on, so nuance costs you nothing at
+merge time. `claims_relied_on` lists the empirical claims the idea would die
+without — that list is what the falsification pass aims at.
+
+## Stress craft (STRESS)
+
+For a claim review, separate **supported**, **refuted**, and **unknown**.
+An observation needs a source and what it actually shows. Agreement across
+ideators is not corroboration; a missing source is not a refutation. Explain
+the consequence for each dependent idea. Preserve failed originals when a
+corrected idea is introduced so the decision can show what changed.
+
+- **pre-mortem** — "it is six months later and this failed — why?"; past-tense
+  framing unlocks concrete failure stories that critique-framing suppresses.
+- **consider-the-opposite** — argue the reverse position seriously once.
+- **disconfirmation** — actively try to kill the front-runner; confirmation
+  accumulates by default, disconfirmation only happens on purpose.
+- Roast moves: sharpest cut first; subtract weak scope before adding caveats.
+- Dissent must cite the observation, not the role ("the wire log shows X"
+  beats "as security, I worry").
+- The pass leaves a trace: fill the record's `pre_mortem`,
+  `disconfirmation_vs_frontrunner`, and `outcome_change` lines — what the
+  stress changed, or exactly why it changed nothing. A stress pass with no
+  recorded verdict is indistinguishable from one that never ran.
+
+## Judge craft (CONVERGE)
+
+- **NUF**: New (does this exist already?), Useful (does it move the actual
+  problem?), Feasible (buildable by this team at this scale?) — 1–10 each,
+  using the full range: a pool judged 4-to-7 is a judge refusing to judge.
+- N is unanchored until it names its neighbor: every finalist carries
+  `nearest existing: <the closest thing that already exists>` and what this
+  beats it on — "none found locally" must be said out loud, never implied.
+- Every finalist carries `cheapest test:` — the one-week falsification a
+  person could start Monday; an idea with no cheap test is a bet, and say so.
+- Write the reason before the score; a score written first back-justifies.
+- Note the asymmetry: reversible? cheap to test? A 6 that's free to try can
+  beat an 8 that's a one-way door.
+- Tag your competence: `inside | edge | needs-evidence` — a judge outside
+  their circle flags it rather than bluffing calibration.
+
+## The record's face (front matter)
+
+The front matter is the first — often the only — thing a decider reads, so it
+opens with `recommendation` and carries no machinery commentary. The
+recommendation is opinionated and tight, in this shape and at most ~60 words:
+
+```text
+<Winner, by name>. <Why, one clause>. Bet: <the single load-bearing
+assumption>. Test: <the falsifier a person could start Monday>.
+```
+
+No hedging words (maybe, could, consider, might) — a run that cannot pick
+says so as its position ("the honest answer is nothing, because <reason>"),
+which is a stance, not a hedge. A menu is not a recommendation; options
+belong in `# The call`, ranked. `foo ideate check` refuses a record whose
+recommendation is still null. Everything below `decider` in the front matter
+is machinery (ids, contract version, the measured meter) — never add
+explainer comments to it; the contract and this skill teach, the record
+states.
+
+## Useful decisions
+
+The rendered decision is the human deliverable. It explains what the proposed
+mechanism changes for the user, what it retains or challenges from named prior
+research and ideas, and why it wins against the best alternative. Give each
+alternative a condition under which the user should choose it.
+
+In every ranking table, put `What this is` immediately after `Direction`.
+Use one plain-English sentence saying who does or receives what. Describe the
+idea itself before explaining why it ranks here; a renamed title or a phrase
+such as “improves efficiency” does not explain the idea. Include the same
+description for excluded directions so readers understand what was rejected.
+
+Make the first experiment executable: an action, a time or cost bound, an
+observable pass criterion, and a stop/change-direction criterion. Treat this
+as a test to propose, not permission to run it. Preserve unresolved empirical
+claims as explicit assumptions. If the evidence kills every candidate,
+recommend doing nothing or gathering the missing evidence; do not manufacture
+a winner to fill the form. Label scope cuts as proposals for the human; evidence-based exclusions remain
+ineligible. Keep excluded directions in the backing rank with their refutation
+and no score. The conductor owns the exact record and rendering
+schema, including the size budget and validation commands.
+
+## Judgment ledger
+
+Carry claim, evidence, inference, confidence, dissent through the run. The
+diversity meter and minority report are ledger entries, not decoration — a
+homogeneous pool or vanished dissent is a finding to surface, never smooth
+over. User-only decisions: the pick, the not-doing list, any scope call.
+
+The meter is computed, never counted by reading: run
+`foo ideate diversity <record> --write` over the stored objects. A number you
+reached by eye must say `diversity_method: estimated` — every downstream reader
+treats the meter as measured, so an eyeballed one is worse than none.
+
+## Usage
+
+```text
+foo ideate "a CLI that reviews my dependencies for supply-chain risk"
+foo ideate "make onboarding not suck" --depth deep
+foo ideate context "make onboarding not suck" --json
+foo ideate check ideas/2026-07-06-dep-risk.md --before-build-on
+foo ideate diversity ideas/2026-07-06-dep-risk.md --write
+foo ideate check ideas/2026-07-06-dep-risk.md
+foo ideate result ideas/2026-07-06-dep-risk.md
+Use $ideate on this idea: <idea>
+```
+
+Run the CLI first; it emits the contract this craft plugs into.
+
+## Error Handling
+
+- **No idea text** — ask for the idea in one sentence; never invent a
+  placeholder.
+- **Contract missing** (skill invoked without a `foo ideate` run) — run the
+  CLI first; do not improvise phases or cells from memory.
+- **Asked to auto-pick or auto-ship** — refuse: the Decider is human (D6);
+  offer the shortlist and the `foo spec` hand-off instead.
+- **Procedural failures** (homogeneous pool, judge inconsistency, missing
+  dissent, budget exhaustion) — the emitted contract carries the recovery
+  rules; follow it, and never bypass `foo ideate check` by hand-editing the
+  verdict.
+- **No prior-work match** — local discovery is lexical and bounded. Inspect
+  related terminology and supplied references; record the search gap honestly.
+- **Unresolved claim** — record unknown, retain its uncertainty in the decision,
+  and propose a test. A record check proves coverage and consistency, not facts.