@sellable/mcp 0.1.136 → 0.1.138

package/skills/interview/references/question-bank.md

@@ -0,0 +1,123 @@
+# Question Bank
+
+Use these questions selectively. Start with source prefill, then ask only for
+gaps that matter for durable memory.
+
+## Core Categories
+
+### Identity
+
+- What shaped your judgment that people would not infer from your resume?
+- What do people misunderstand about how you think or work?
+- What do you refuse to pretend, even when it would sound more polished?
+
+### Voice Sample Inventory
+
+- What voice samples exist: posts, emails, comments, docs, call transcripts,
+  meeting transcripts, voice memo transcripts, or sales replies?
+- Who is the target speaker in each sample?
+- Is the sample representative, stale, ghostwritten, AI-assisted, heavily
+  edited, natural speech, or polished writing?
+- What context labels, approximate word count, audience, and sample quality
+  should be recorded?
+
+### Topic Transcript Inventory
+
+- What topics need durable opinion coverage?
+- Which transcript entries should carry topic tags and links back to derived
+  memory files?
+- Which topics are underdeveloped enough to require a follow-up interview?
+
+### Reusable Answer Capture
+
+- Is this answer specific enough to save?
+- What context applies: investor, accelerator, LinkedIn, outbound, website,
+  sales reply, technical docs, internal strategy?
+- Is the wording public-safe, private, or safe only after rewrite?
+- What source proves it?
+- Which derived memory files should update: `answer-bank.md`,
+  `proof-ledger.md`, `story-bank.md`, `decision-rules.md`, `my-company.md`, or
+  `context-modes.md`?
+
+### Taste
+
+- What writing makes you cringe?
+- What do you type and then delete?
+- What examples feel unmistakably right, and why?
+
+### Company Truth
+
+- What does the product do today?
+- What changed recently?
+- What sounds impressive but is not safe to claim?
+- What buyer/customer language is safest to reuse?
+
+### Proof Hygiene
+
+- Which metrics are verified, directional, private, stale, or forbidden?
+- What confidence and source should each claim carry?
+- What must the agent never say?
+
+### Stories
+
+- What customer conversation, failure, expensive lesson, or origin moment
+  explains the product better than polished positioning?
+
+### Context Modes
+
+- How should the user sound differently in comments, posts, cold outbound,
+  investor answers, sales replies, technical docs, sensitive situations, and
+  website copy?
+
+### Anti-AI
+
+- Which AI tells are hard bans, context-only bans, user-authentic exceptions,
+  or allowed factual contrasts?
+
+### Reference Curation
+
+- Which personal LinkedIn posts, admired examples, applications, and sales-copy
+  examples should be copied, distilled, indexed only, or banned?
+
+## Privacy And Security
+
+Classify voice files, answer-bank entries, and transcripts by sensitivity.
+Before creating shareable/team-facing versions, ask whether private stories,
+names, metrics, customers, or transcript details should be removed. Record
+shareability and public/private reuse on saved answers.
+
+## Source-Mapped Questions
+
+### Ruben Text-File Method
+
+- What do people misunderstand?
+- What do you refuse to pretend?
+- What are your signature tells and phrase bank?
+- What are hard refusals, taste loves/disgusts, productive contradictions,
+  golden examples, and do-not-infer boundaries?
+
+### Ruben Anti-AI Method
+
+- Which AI tells are hard bans?
+- Which are context-only bans?
+- Which patterns are user-authentic exceptions?
+- Which contrasts are allowed factual contrasts rather than fake reframes?
+- What anti-overfitting risk would make the writing worse?
+
+### KMO Voice-Sample Method
+
+- What sample inventory exists?
+- Who is the target speaker?
+- Which samples are representative vs stale/ghostwritten/AI-assisted?
+- What frequent and avoided words, signature phrases/transitions, sentence
+  structure, metaphor/analogy habits, argument structure, anecdotes, audience
+  engagement, and technical vs nontechnical variation appear?
+
+### Sellable-Specific Company/Proof Needs
+
+- What is current product truth?
+- What stale positioning should be retired?
+- What proof source/confidence is safe?
+- What is public, private, or needs rewrite?
+- What buyer/customer language should be reused?
+- What claims must the agent never make?

package/skills/interview/references/reference-curation.md

@@ -0,0 +1,68 @@
+# Reference Curation
+
+Curated references give future writing skills concrete examples without forcing
+them to reread every raw archive. Use copied/distilled examples when the example
+itself is useful context; use indexed-only entries when the source is large,
+sensitive, external, or only useful for provenance. Every folder uses INDEX.md manifests so examples remain source-mapped and idempotent.
+
+## Folders
+
+- `.sellable/configs/core/references/linkedin-posts/` - personal best posts,
+  comments, hooks, and thread patterns.
+- `.sellable/configs/core/references/inspiration/` - admired outside examples,
+  anti-examples, voice-capture sources, and pattern notes.
+- `.sellable/configs/core/references/applications/` - accelerator, investor,
+  founder-profile, and answer-pack references.
+- `.sellable/configs/core/references/sales-copy/` - outbound, reply, website,
+  deck, objection, proof, and customer-language examples.
+
+Each folder should have an `INDEX.md`. A copied/distilled example should live
+near its index. Keep sensitive originals out of shareable folders unless the
+user approves.
+
+## Manifest Fields
+
+Every `INDEX.md` entry must include:
+
+- title
+- source path/URL
+- copied file path or `indexed only`
+- tags
+- why it matters
+- last reviewed
+- usage contexts
+- privacy/sensitivity
+- allowed use: quote, imitate structure, summarize, private only, or banned
+
+## Copy Or Index
+
+Copy or distill when:
+
+- the exact language is a reusable voice example;
+- the structure should be imitated;
+- the claim/proof/story is likely to be used repeatedly;
+- the source may disappear or is hard to fetch later.
+
+Index only when:
+
+- the source is private, long, or legally sensitive;
+- the example is useful only as provenance;
+- a short note is enough to preserve why it matters;
+- copying would expose private customers, names, or metrics.
+
+## Duplicate Keys
+
+Use normalized URL/source path/title plus optional content hash as duplicate
+keys. If a source appears again, update its `last reviewed` and usage contexts
+instead of creating a second entry.
+
+## Review Notes
+
+Every reference entry should say why it matters, which contexts can use it, and
+whether the example is safe to quote, imitate structurally, summarize, or keep
+private. If an example is an anti-example, label the pattern to avoid.
+
+Personal LinkedIn posts and inspiration examples deserve especially clear
+notes: explain whether the lesson is voice, hook shape, proof handling,
+argument structure, audience fit, or a pattern to avoid. Do not mix admiration
+with permission to copy; many inspiration examples should be structural only.

package/skills/interview/references/voice-capture-method.md

@@ -0,0 +1,62 @@
+# Voice Capture Method
+
+This is the single operating model for the core interview. Ruben/KMO/source
+materials are provenance; do not expose them as separate user-facing workflows.
+
+## Operating Loop
+
+1. Prefill from existing `.sellable/configs`, docs, posts, transcripts, emails,
+   call notes, and user-provided samples.
+2. Do evidence collection: collect source path/URL, date, author/speaker,
+   context, audience, privacy, and whether the sample is natural speech, edited
+   writing, ghostwritten, or AI-assisted.
+3. Do speaker filtering: for multi-speaker transcripts, identify the target
+   speaker(s), exclude others from voice-pattern extraction, and keep other
+   speakers only as context or contrast.
+4. Do observed pattern extraction before asking self-description questions:
+   words, avoided words, sentence length, rhythm, transitions, examples,
+   argument shape, story habits, proof style, and recurring refusals.
+5. Run the gap interview: ask only for missing identity, company truth, proof,
+   taste, context modes, transcript topics, answer-bank candidates, and
+   reference curation.
+6. Do durable compile: raw archive first, then stable core files, answer bank,
+   transcript library, proof/story/rule updates, and reference manifests.
+7. Audit/correct: run the anti-AI audit, show compact candidate memory, ask for
+   correction, and preserve rejected or private material in the right place.
+8. Rerun idempotently: every future run adds, corrects, merges, or retires by
+   stable keys without wiping manual edits.
+
+## Keep
+
+- Source prefill before the first question.
+- Natural samples and transcripts as evidence, with target-speaker filtering.
+- Observed language patterns over generic self-description.
+- Correction loops: "here is what I think I know" before a questionnaire.
+- Raw archives separate from curated transcript entries.
+- Approved reusable answers separate from raw answers.
+- Compact compiled memory that changes future behavior.
+- Anti-AI style rules with user-authentic exceptions.
+- Privacy and sensitivity labels for samples, transcripts, and answer-bank
+  entries.
+- Idempotent reruns with stable keys, generated-section markers, and manual
+  edit preservation.
+
+## Drop
+
+- Do not require PDF export.
+- Do not require custom GPT setup.
+- Do not rely on time-sensitive model claims such as "Claude is best."
+- Do not require exact sample-size gates. A large corpus helps, but it is not a
+  blocking prerequisite.
+- Do not do theatrical impersonation or invent personality quirks.
+- Do not optimize for AI-detector avoidance at the expense of accurate,
+  natural, user-authentic writing.
+- Do not use giant questionnaires as the default. The 100-question version is
+  opt-in deep mode.
+
+## Output Discipline
+
+The interviewer should make source-backed claims about voice. If evidence is
+thin, label the rule as tentative and ask for a sample or correction. Never
+promote a phrase to a signature tell just because it appeared once in an edited
+or ghostwritten sample.