@sellable/mcp 0.1.257 → 0.1.259

package/skills/create-post/SKILL.md

@@ -84,7 +84,6 @@ Before drafting, load all required assets with `mcp__sellable__get_subskill_asse
 3. `subskillName: "create-post", assetPath: "references/premise-development.md"`
 4. `subskillName: "create-post", assetPath: "references/post-validation.md"`
 5. `subskillName: "create-post", assetPath: "references/gold-standard-post-pack.md"`
-6. `subskillName: "create-post", assetPath: "references/linkedin-preview-rendering.md"`
 
 If any required asset is missing, unreadable, truncated without continuation, or internally inconsistent, return:
 
@@ -305,8 +304,7 @@ If local idea capture succeeds but auth/workspace is missing, keep the idea and
 
 ## Step 1: Hook Research
 
-Use `references/hook-research-playbook.md` and
-`references/linkedin-preview-rendering.md`.
+Use `references/hook-research-playbook.md`.
 
 If the host supports background agents, delegate the search/fetch/autopsy work
 to one bounded `research-worker` before hook candidate generation. The worker
@@ -321,10 +319,9 @@ The research worker must return a compact packet only:
 - full adapted hook blocks
 - market belief map: resonating ideas, implicit beliefs, audience wants, resentments, fears, and credible controversy angles
 - premise inputs: real scenes, observed tensions, reader value openings, and proof gaps
-- reach-normalized signal notes, including follower-band fit when available
 - exact phrase patterns and sentence shapes
 - body structures and exact body language moves
-- rendered mobile and desktop preview records
+- preview measurements
 - track-person and gold-standard recommendations
 - blocked states or confidence gaps
 
@@ -335,20 +332,18 @@ the exact extracted phrase shapes.
 Default flow:
 
 1. Convert the idea into 3-8 search keywords.
-2. Call `mcp__sellable__search_engagement_posts` with an explicit multi-month window. Default to `maxAgeDays: 120`, tightening to 30-60 days only when the topic is trend-sensitive. When the user gives a follower range, pass it as `targetFollowerMin` and `targetFollowerMax`; for example, "8k-20k followers" means `targetFollowerMin: 8000` and `targetFollowerMax: 20000`.
-3. Shortlist posts by topic fit, rendered hook strength, content pattern replicability, creator repeat evidence, weighted engagement quality, and reach-normalized signal quality. Do not sort by raw engagement alone, and do not let the numeric reach score choose the final hook by itself.
-4. Because search results may only include previews, call `mcp__sellable__fetch_linkedin_posts` for shortlisted authors/profile URLs and match recent posts by URL/activity ID when full text is needed. If a promising source is missing follower count and the user requested reach normalization, call `mcp__sellable__fetch_linkedin_profile` on a bounded shortlist before treating the source as a keeper.
+2. Call `mcp__sellable__search_engagement_posts` with an explicit multi-month window. Default to `maxAgeDays: 120`, tightening to 30-60 days only when the topic is trend-sensitive.
+3. Shortlist high-engagement posts by topic fit, hook strength, creator repeat evidence, and weighted engagement quality.
+4. Because search results may only include previews, call `mcp__sellable__fetch_linkedin_posts` for shortlisted authors/profile URLs and match recent posts by URL/activity ID when full text is needed.
 5. If full text cannot be matched, record `full_text_unavailable` and use only the preview. Do not invent missing body details.
-6. Weigh shares/reposts above comments, comments above reactions, and reactions as weak reach unless paired with stronger signals. Normalize engagement against follower count when available: record `authorFollowerCount`, `targetFollowerBand`, `followerBandFit`, `engagementPer1kFollowers`, `weightedEngagementPer1kFollowers`, `reachPenaltyMultiplier`, `reachAdjustedScore`, and `baselineLift` when repeated creator posts make it possible. If follower count is unavailable, record `follower_count_unavailable`; if shares/reposts are unavailable, record `repost_data_unavailable`.
+6. Weigh shares/reposts above comments, comments above reactions, and reactions as weak reach unless paired with stronger signals. If shares/reposts are unavailable, record `repost_data_unavailable`.
 7. Penalize lead-magnet or giveaway mechanics unless the user explicitly asks for a lead magnet post.
 8. Build a market belief map before hook generation: what the space is rewarding, what the audience implicitly believes, what they want permission to say, what they resent or fear, what they will argue with, and which controversial angle the user can credibly own. If the user's raw idea is internally coherent but not attached to a live market tension, do not draft from the internal idea alone; present stronger directions or rewrite the draft angle around the external tension.
 9. Extract premise inputs: real story/scene possibilities, observed tensions, useful reader takeaways, and proof gaps. A good hook cannot rescue a premise with no value.
 10. For story posts, extract the story mechanism that made the post work, not just the first line.
 11. Extract hook structures plus specific reusable words, phrases, sentence
    shapes, transitions, and body language patterns.
-12. Render each kept source hook and adapted hook block through the LinkedIn
-   preview rendering contract. Character counts alone are not enough.
-13. Save the research with `mcp__sellable__save_hook_research`.
+12. Save the research with `mcp__sellable__save_hook_research`.
 
 Record provenance:
 
@@ -357,26 +352,18 @@ Record provenance:
 - search window
 - source post URLs
 - authors/profile URLs
-- author follower counts and target follower-band fit when available
 - engagement totals and available likes/comments/shares breakdown
-- reach-normalized scoring fields and confidence notes
 - creator repeat evidence
 - lead-magnet or engagement-bait penalties
 - story mechanism when relevant
 - full-text match status
-- source hook rendered preview records and whether they came from full text,
-  authenticated LinkedIn screenshots, or a search preview
+- source hook preview measurements and whether they came from full text or a search preview
 - selected hook patterns
 - market belief map and selected controversy
 - premise cards and selected premise
 - exact phrase patterns and sentence shapes
 - body structures and body language patterns
 - why each pattern fits the user's idea and voice
-- `whyTheHookCarries`: why the selected hook works from the words, tension, and
-  content pattern independent of the source creator's reach
-- `whyTheReachEvidenceIsTrustworthy`: follower-band fit, reach-adjusted score,
-  baseline lift, share/comment quality, or why a large-account example is only
-  secondary pattern evidence
 
 ## Step 1.5: Research Learning Report
 
@@ -402,9 +389,6 @@ Research status:
 - keywords:
 - full-text coverage:
 - repost/share data:
-- target follower band:
-- reach-normalized winners:
-- big-account examples used only as secondary pattern evidence:
 
 Best source examples:
 1. author, URL, engagement, why kept, why not copied
@@ -425,10 +409,7 @@ Premise cards:
 Hook patterns learned:
 1. full adapted hook block
    - source mechanism:
-   - rendered preview:
-     - mobile visible block:
-     - desktop visible block:
-     - verdict:
+   - preview budget:
    - internal question:
    - why it fits / why it does not:
 
@@ -503,15 +484,6 @@ Each hook must include:
 - reader value implied
 - source hook pattern
 - why it fits this idea
-- `renderedPreview` using `references/linkedin-preview-rendering.md`
-- `mobileRenderedPreviewBlock`
-- `desktopRenderedPreviewBlock`
-- `mobileRenderedLines`
-- `desktopRenderedLines`
-- `firstScreenPromise`
-- `corePainProofOrCuriosityVisibleMobile`
-- `corePointVisibleMobile`
-- `pointAfterMobileClamp`
 - `charCount`
 - `charCountIncludingNewlines`
 - `firstLineChars`
@@ -532,39 +504,22 @@ Each hook must include:
 
 Do not copy source wording. Copy only the structure.
 
-Render every candidate before scoring it. The LinkedIn preview rendering
-contract is the gate; character counts are secondary diagnostics only. Use the
-deterministic CSS contract when no authenticated LinkedIn screenshot is
-available:
-
-- mobile text width: `308px`
-- desktop text width: `582px`
-- font size: `14px`
-- line height: `21px`
-- white space: `pre-wrap`
-- overflow wrap: `break-word`
-- review clamp: first 3 rendered text lines
-
-Use the rendered gates from `references/linkedin-preview-rendering.md`:
-
-- `pass`: the mobile rendered preview shows the pain, proof, or curiosity by
-  the end of the first 3 rendered lines, and the core point is understandable
-  without opening "see more".
-- `warn`: the mobile rendered preview creates useful curiosity but the core
-  point is slightly softened by wrapping, blank-line rhythm, or one missing
-  context word. A compact fallback is required.
-- `fail`: the hook's real point appears after the first 3 mobile rendered
-  lines, the first rendered line is generic setup, blank lines consume the
-  preview before the reader sees the point, or desktop fit is the only reason it
-  looks good.
+Use this conservative mobile-first LinkedIn preview gate. LinkedIn does not
+publish exact "see more" cutoff rules, and rendering varies by device, app
+version, font, media, and line break. These are v1 safety budgets, not claims
+about an official LinkedIn limit:
+
+- `pass`: hook is <= 110 chars including newlines, every nonblank line is <= 45 chars, and the hook's core point lands before likely truncation.
+- `warn`: hook is 111-140 chars including newlines, any nonblank line is 46-55 chars, or blank lines create visual-line risk. Blank lines are allowed, but they count as physical lines.
+- `fail`: hook is > 140 chars including newlines, any nonblank line is > 55 chars, or the hook's point depends on text after likely truncation.
 
 Desktop preview usually has more room. Still record `desktopPreviewFit`, but
 never let desktop fit compensate for a mobile `fail`.
 
-If a hook's point depends on text after the rendered mobile preview clamp,
-rewrite it before selecting it. A selected hook may carry a `warn` only when
-the warning is explicit and the validation receipt includes a compact fallback.
-A hook with no rendered mobile and desktop preview record cannot be selected.
+If a hook's point depends on text after the likely preview, rewrite it before
+selecting it. A selected hook may carry a `warn` only when the warning is about
+intentional blank-line rhythm or a slight line-length overage; include a compact
+fallback in the validation receipt.
 
 ## Step 3: Draft
 
@@ -598,7 +553,6 @@ Every saved draft needs a validation receipt with:
 - story/proof files consulted
 - gold standards consulted
 - LinkedIn preview audit
-- rendered mobile and desktop hook preview blocks
 - premise/value audit findings
 - simplifier/concrete-language audit findings
 - voice audit findings

package/skills/create-post/references/gold-standard-post-pack.md

@@ -119,8 +119,6 @@ Show compact candidate cards. Include:
 - why it might belong in the pack
 - hook mechanism
 - hook preview budget status and measurement basis
-- rendered mobile preview block and verdict
-- rendered desktop preview block and verdict
 - content/body mechanism
 - rhythm notes
 - sentence-structure notes
@@ -196,15 +194,6 @@ Tags:
 - longest nonblank line chars:
 - blank-line visual risk:
 - source text basis:
-- render basis:
-- css contract version:
-- mobile rendered preview block:
-- desktop rendered preview block:
-- mobile rendered lines:
-- desktop rendered lines:
-- core pain/proof/curiosity visible on mobile:
-- core point visible on mobile:
-- point after mobile clamp:
 - preview budget status:
 - mobile preview fit:
 - desktop preview fit:

package/skills/create-post/references/hook-research-playbook.md

@@ -28,11 +28,10 @@ Worker owns:
 - tracked-person post fetches
 - full-text matching by URL/activity ID
 - duplicate removal
-- follower-band and reach-normalized signal scoring
 - lead-magnet, giveaway, engagement-bait, and off-voice filtering
 - market belief mapping across kept and rejected examples
 - premise input extraction: real scenes, observed tensions, reader value, proof gaps
-- rendered hook opening measurement
+- hook opening measurement
 - exact phrase-pattern extraction
 - body-structure extraction
 - rejected-example notes
@@ -59,8 +58,7 @@ Research packet:
 - exact phrase patterns: max 20
 - body patterns: max 8
 - source URLs and author profile URLs
-- reach-normalized signal notes
-- rendered preview records
+- preview measurements
 - confidence gaps
 - save recommendations
 ```
@@ -76,13 +74,6 @@ Use `mcp__sellable__search_engagement_posts` with practical constraints:
 
 - explicit `maxAgeDays: 120` by default so research covers the past 30-120 days, not only this week's posts
 - tighten to 30-60 days for trend-sensitive topics; widen only when the topic has low volume
-- when the user gives a follower range, pass it as `targetFollowerMin` and
-  `targetFollowerMax` so the tool can expose reach-normalized signals and
-  prioritize comparable creators
-- if a source hook looks promising but follower count is missing, use
-  `mcp__sellable__fetch_linkedin_profile` on a bounded shortlist before calling
-  the source a keeper; if profile fetch is unavailable or too slow, mark
-  `follower_count_unavailable`
 - high enough engagement to matter
 - narrow enough to match the idea
 - enough result depth to see whether a creator has repeated winners, not just one viral outlier
@@ -91,19 +82,14 @@ Record every keyword, filter, search window, page, and result count.
 
 ## Weighted Signals
 
-Rank source posts with a weighted signal view. Do not sort by total engagement
-alone. Use the numeric score to decide which sources deserve study; do not let
-the number choose the final hook by itself.
+Rank source posts with a weighted signal view. Do not sort by total engagement alone.
 
 Use this scoring shape:
 
 ```text
 hook_score =
   topic_fit
-  + rendered_mobile_preview_strength
   + hook_clarity
-  + content_pattern_replicability
-  + reach_adjusted_evidence
   + creator_repeat_success
   + repost_share_strength
   + comment_quality
@@ -121,135 +107,6 @@ Guidance:
 - If share or repost data is unavailable, record `repost_data_unavailable` instead of inventing it.
 - Prefer creators with repeated high-performing posts in the same lane over one-off viral posts.
 
-## Reach-Normalized Hook Scoring
-
-Raw engagement is not the clearest signal when source creators have wildly
-different audience sizes. A post from a 100k+ follower creator can win on
-distribution even when the hook is ordinary. A post from a creator near the
-user's follower range that overperforms is a clearer signal that LinkedIn and
-readers rewarded the hook/premise.
-
-This calculation is a source-quality filter, not the final creative decision.
-After reach is controlled, the LLM still chooses the hooks to steal based on the
-visible hook, rendered first-screen promise, content pattern, premise fit, and
-whether the idea can carry without the source creator's distribution.
-
-When the user gives a target range, default to that range. For example, if the
-user says "8k to 20k followers," call `mcp__sellable__search_engagement_posts`
-with:
-
-```text
-targetFollowerMin: 8000
-targetFollowerMax: 20000
-```
-
-If the user gives no range, use the user's known follower count from memory or
-ask only when the decision depends on it. If no follower count is available, use
-raw engagement as a fallback and mark `target_follower_band_unknown`.
-
-For every shortlisted source post, record:
-
-- `authorFollowerCount`
-- `targetFollowerBand`
-- `followerBandFit`: `in_target_band`, `below_target_band`,
-  `above_target_band`, or `unknown`
-- `engagementPer1kFollowers`
-- `weightedEngagementPer1kFollowers`
-- `reachAdjustedScore`
-- `creatorBaselineMedianEngagement` when repeated posts from that creator are
-  available
-- `baselineLift`: this post's engagement divided by the creator's recent median
-  engagement in the same lane, or `creator_baseline_unavailable`
-- `confidence`: `high`, `medium`, or `low`
-- `normalizationNotes`
-
-Use this Harvest-compatible calculation when follower counts are available:
-
-```text
-weightedEngagement =
-  likes
-  + (comments * 4)
-  + (shares * 12)
-
-engagementPer1kFollowers =
-  totalEngagement / authorFollowerCount * 1000
-
-weightedEngagementPer1kFollowers =
-  weightedEngagement / authorFollowerCount * 1000
-
-reachPenaltyMultiplier =
-  1.00 when in target follower band
-  0.75 when below target band
-  0.65 when above target band but <= 2x target max
-  0.35 when above target band but <= 5x target max
-  0.15 when above target band and > 5x target max
-  0.40 when follower count is unknown
-
-reachAdjustedScore =
-  weightedEngagementPer1kFollowers * reachPenaltyMultiplier
-```
-
-When repeated posts from the same creator are available, calculate:
-
-```text
-creatorBaselineMedianEngagement =
-  median weightedEngagement across recent same-lane posts
-
-baselineLift =
-  post weightedEngagement / creatorBaselineMedianEngagement
-```
-
-Use this scoring shape when deciding which hooks to study:
-
-```text
-hook_score =
-  topic_fit
-  + rendered_mobile_preview_strength
-  + hook_clarity
-  + same_follower_band_bonus
-  + weighted_engagement_per_1k_followers
-  + creator_baseline_lift
-  + repost_share_strength
-  + substantive_comment_quality
-  + creator_repeat_success
-  - celebrity_reach_penalty
-  - tiny_sample_penalty
-  - lead_magnet_penalty
-  - engagement_bait_penalty
-  - off_voice_penalty
-```
-
-Rules:
-
-- Prefer in-band creators when the hook is strong and the engagement is real.
-- Use near-band creators as secondary evidence.
-- Penalize creators far above the target range, especially accounts above 5x
-  the target max or above 100k followers when the target range is 8k-20k.
-- Do not automatically reject large-account posts; they can still teach body
-  structure, topic heat, or category language. Do not treat them as primary hook
-  proof unless they also overperform their creator baseline.
-- Large-account hooks can be selected only when the hook carries without the
-  creator: the rendered mobile opening is strong, the premise is reusable by the
-  user, the body pattern does not depend on celebrity authority, and the source
-  either beats baseline or has unusually high reach-adjusted quality after the
-  penalty.
-- Do not overvalue tiny-account anomalies. Very high engagement per follower
-  with low absolute engagement is interesting but lower confidence.
-- When follower data is unavailable, do not invent it. Mark
-  `follower_count_unavailable`, lower confidence, and rely more on creator repeat
-  evidence, repost/share strength, comment quality, and rendered preview quality.
-- A source hook is strongest when it is in/near the target follower band, has
-  real absolute engagement, beats the creator's apparent baseline, has
-  share/comment quality, and passes rendered mobile preview.
-
-Final hook selection must explain both:
-
-- `whyTheHookCarries`: the visible words, specificity, tension, and first-screen
-  promise that work independent of reach
-- `whyTheReachEvidenceIsTrustworthy`: follower-band fit, reach-adjusted score,
-  baseline lift, share/comment quality, or why a large-account example is only
-  secondary pattern evidence
-
 Penalize lead-magnet and engagement-bait mechanics unless the user explicitly asks for that style:
 
 - `comment "template"` / `comment "guide"` / `comment "playbook"`
@@ -341,48 +198,24 @@ Measure the visible opening for every shortlisted source post before extracting
 the hook pattern. This makes the study useful for LinkedIn, not just generally
 "good writing."
 
-Use `references/linkedin-preview-rendering.md`. LinkedIn does not publish exact
-"see more" cutoff rules. Character counts are diagnostics only. A source post
-or generated hook is not properly studied until it has a rendered mobile and
-desktop preview record.
-
-Default deterministic renderer for unpublished drafts and research comparison:
+LinkedIn does not publish exact "see more" cutoff rules. Treat these as
+conservative v1 planning budgets:
 
-```text
-cssContractVersion: linkedin-preview-rendering/v1
-font size: 14px
-line height: 21px
-white space: pre-wrap
-overflow wrap: break-word
-mobile text width: 308px
-desktop text width: 582px
-review clamp: first 3 rendered text lines
-```
+- `pass`: opening hook is <= 110 chars including newlines, every nonblank line
+  is <= 45 chars, and the hook's core point lands before likely truncation.
+- `warn`: opening hook is 111-140 chars including newlines, any nonblank line is
+  46-55 chars, or blank lines create visual-line risk. Blank lines are allowed,
+  but they count as physical lines.
+- `fail`: opening hook is > 140 chars including newlines, any nonblank line is
+  > 55 chars, or the hook's core point depends on text after likely truncation.
 
-Authenticated LinkedIn feed/composer screenshots are stronger evidence when
-available. If a screenshot is used, still record the wrapped lines and verdicts
-below so another agent can compare hooks without re-opening LinkedIn.
-
-Desktop preview has more room, but never let desktop fit compensate for a
-mobile `fail`.
+Desktop preview has more room, so record it separately, but never let desktop
+fit compensate for a mobile `fail`.
 
 For each source, record:
 
 - `sourceTextBasis`: `full_text`, `search_preview`, or `manual_user_source`
 - `openingTextUsed`
-- `renderedPreview`
-- `cssContractVersion`
-- `mobileRenderedPreviewBlock`
-- `desktopRenderedPreviewBlock`
-- `mobileRenderedLines`
-- `desktopRenderedLines`
-- `mobileRenderedLineCount`
-- `desktopRenderedLineCount`
-- `corePainProofOrCuriosityVisibleMobile`
-- `corePainProofOrCuriosityVisibleDesktop`
-- `corePointVisibleMobile`
-- `corePointVisibleDesktop`
-- `pointAfterMobileClamp`
 - `charCountIncludingNewlines`
 - `physicalLineCount`
 - `contentLineCount`
@@ -395,25 +228,11 @@ For each source, record:
 - `desktopPreviewBudget`: `pass`, `warn`, or `fail`
 - `blankLineVisualRisk`
 - `corePointBeforeLikelyTruncation`
-- `renderedPreviewVerdict`: `pass`, `warn`, or `fail`
 
 If only a search preview is available, do not pretend the opening is complete.
 Record `sourceTextBasis: search_preview` and lower confidence when the hook
 appears cut off or body context is unavailable.
 
-Pass/warn/fail is based on rendered output:
-
-- `pass`: the mobile rendered preview shows the pain, proof, or curiosity by
-  the end of the first 3 rendered lines, and the core point is understandable
-  without opening "see more".
-- `warn`: the mobile rendered preview creates useful curiosity but the core
-  point is slightly softened by wrapping, blank-line rhythm, or one missing
-  context word. A compact fallback is required.
-- `fail`: the hook's real point appears after the first 3 mobile rendered
-  lines, the first rendered line is generic setup, blank lines consume the
-  preview before the reader sees the point, or desktop fit is the only reason it
-  looks good.
-
 ## Hook Extraction
 
 Extract structure and reusable language patterns, not copied prose. The goal is
@@ -426,13 +245,10 @@ For each shortlisted source post, record:
 - URL
 - author
 - author profile URL when available
-- author follower count and follower-band fit when available
 - engagement totals and available likes/comments/shares breakdown
-- reach-normalized metrics: engagement per 1k followers, weighted engagement
-  per 1k followers, reach-adjusted score, baseline lift, and confidence
 - creator repeat evidence
 - visible hook text or preview
-- rendered preview fields from the section above
+- opening preview measurement fields from the section above
 - hook mechanism
 - exact hook language patterns: reusable words, phrase shapes, contrast forms,
   sentence shapes, and transition moves
@@ -442,16 +258,10 @@ For each shortlisted source post, record:
 - proof/story dependency
 - lead magnet or engagement bait penalty
 - weighted signal notes
-- reach-normalized signal notes
 - replicability score
 - track person recommendation: `yes`, `no`, or `ask_user`
 - tracking reason when recommended
 
-Do not call a source hook a keeper because the full text is strong if the
-rendered mobile opening is weak. Study what a LinkedIn reader sees before
-"see more"; the body structure can still be useful, but the hook should not be
-copied into the hook pattern set.
-
 ## Specific Language Extraction
 
 For each keeper, extract the source's specific language mechanics in this

package/skills/create-post/references/post-file-contract.md

@@ -59,19 +59,10 @@ Hook research files must preserve:
 - source post URLs
 - author/profile URLs
 - engagement totals
-- author follower counts when available, target follower band, follower-band
-  fit, engagement per 1k followers, weighted engagement per 1k followers,
-  reach penalty multiplier, reach-adjusted score, baseline lift when available,
-  why the hook carries independent of creator reach, and normalization
-  confidence notes
 - full-text availability
 - source hook preview measurements, including text basis, char count including
   newlines, physical/content line counts, longest nonblank line, blank-line
   visual risk, and mobile/desktop preview budget status
-- rendered preview records for every kept source hook and adapted hook block,
-  including literal mobile/desktop preview blocks, rendered line wraps, render
-  basis, CSS contract version, text widths, first-screen promise visibility,
-  core point visibility, and whether the point lands after the mobile clamp
 - extracted hook patterns
 - selected hook basis
 
@@ -92,9 +83,6 @@ Draft files must preserve:
 - draft body
 - validation receipt, including LinkedIn preview pass/warn/fail status and
   compact fallback when the selected hook carries a warning
-- rendered mobile and desktop preview blocks for the selected hook; drafts
-  cannot be ready when this rendered-preview audit is missing or fails mobile
-  visibility
 - status: `draft`, `ready`, `needs_revision`, `published`, or `archived`
 
 Multiple drafts for the same idea are expected. Keep the `ideaId` stable and