Content Skills Voice Audit — 2026-07-07

Follow-up to Voice Guide System Audit 2026-07-07, per the session prompt. Scope: 11 content-creation skills in .claude/skills/ and 5 companion guideline docs, checked against five dimensions: voice loading, reference-don’t-duplicate, Final Voice Pass, per-platform trajectory, and repetition scan.

Status: fixes applied 2026-07-07, same session. Patrick approved Groups 1–7 and Decision D (group-post neutral register → VOICE-GUIDE-UNIFIED v2.9 draft, pending Ksenia). Decision A resolved: hashtags go inside the post content on every platform; the Instagram first-comment convention is retired entirely (Patrick: “we’re never going to post the first comment”). Decision B resolved 2026-08-03: campaign/administrative/bulk emails carry no signature (they speak as Baseworks); personalized emails drafted in a personal voice are signed by whoever is sending — the distinction is now explicit in /create-email. Decision C defaulted to no change (the Teacher-Student section quotes the published teaching principles). All 11 skills, 4 companion docs, the skills index (table rows + detailed sections, plus a previously missing /create-newsletter section), and the unified guide were updated; Asia inbox item added for v2.9.

Documents audited: create-campaign, create-newsletter, create-email, write-blog-post, write-communications-post, write-page-copy, post-to-group, draft-response, create-event-post, update-science-content, session-summary; plus Campaign Creation Guidelines, Blog Post Guidelines, Archive Newsletter Guidelines, Social Media Tags Reference, Communications Guide.

---

Part 1 — Cross-cutting findings

Finding 1: The Final Voice Pass exists in zero skills

WRITING-PROCESS v1.1 (added today) requires one logged voice-only pass after content settles, as the fix for cross-round regression. No skill currently ends with it. Every content-producing skill closes with a content-review or handoff phase; the closest any gets is write-blog-post Phase 4 (“Voice check”), which is a re-read folded into content review, exactly the pattern the audit showed regressing.

Fix (all 11 skills): add one short step at the review/finalization phase that points to “The Final Voice Pass” in WRITING-PROCESS.md, does not restate its content, and requires the pass to be logged in the piece’s Editorial Notes (or, for short-form output like responses and group posts, reported inline to the user).

Finding 2: Voice-loading gaps

Skill	WRITING-PROCESS	Unified guide	Personal guide by author	Verdict
create-campaign	yes	yes	yes (both options)	OK
write-blog-post	yes	yes	yes (both options)	OK
write-communications-post	yes	yes	yes (both options)	OK
write-page-copy	yes	yes	yes (both options)	OK
update-science-content	no (referenced late, in Step 7 only)	yes	Ksenia for articles; pages unified-only	Add WRITING-PROCESS to pre-flight
create-newsletter	no	yes	Patrick only, unconditional	Add WRITING-PROCESS; make personal guide author-conditional
create-email	no	yes	Patrick only, unconditional	Same
draft-response	no	yes	yes (both options)	Add WRITING-PROCESS
session-summary	no	no	no	Gap. Summaries are written in the posting instructor’s voice; the unified guide’s By Context section names session summaries explicitly. The skill relies entirely on _session-summary-guidelines.md. Add unified + posting-instructor guide to Pre-Flight step 3.
post-to-group	no	no	no	Gap. The skill edits and “fixes” body copy (its own §1d checklist) without ever loading the guide the checklist summarizes.
create-event-post	no	no	no	Gap. It drafts event description copy, SEO titles, meta descriptions, OG copy, and alt text with no voice guidance loaded at all.

None of the skills state the read-before-ANY-draft rule (including quick inline chat drafts) now codified in VOICE-GUIDE-PATRICK §Process. The pre-flight ordering mostly covers it, but the loading step in each skill should say “before any drafting, including inline drafts presented in chat,” so a session that skips ahead still hits the rule.

Finding 3: Duplicated rules (the drift mechanism)

Per the audit’s second structural finding, local copies of voice rules go stale. Instances found, by severity:

1. create-campaign §2a “Load cross-referenced conventions from memory” — restates ~10 rules as bullet summaries AND points at Claude’s machine-local memory files as the source. Both halves are the drift mechanism: the summaries will go stale, and the memory files don’t exist on Asia’s machines or the VPS. It also references feedback-facebook-no-linkaway.md, which no longer exists (that rule was retired 2026-04-23) — a live demonstration of the problem. Fix: codify the platform rules into the campaign guidelines (Finding 4), then reduce §2a to pointers at the guideline docs.
2. post-to-group §1d voice compliance table — restates em-dash, exclamation, negative-opener, “land on,” and metaphor rules inline. Keep the check step; replace the table with “run the mechanical scan from the Final Voice Pass (WRITING-PROCESS) / unified guide Avoid list.”
3. create-email “Em dash audit” — full replacement recipe duplicated from unified guide §6a. Keep the audit-and-report step (it’s good process); trim the recipe to a pointer.
4. session-summary Phase 5.2 — same full em-dash recipe duplicated.
5. draft-response Conventions — restates em-dash, contractions, opener (“Thank you for your inquiry”), and fit-framing rules, all of which now live in the unified guide. Trim to pointers.
6. write-blog-post Phase 3 — restates the AI Generation Warning failure mode and three filters at length. Trim to a pointer (the skill already mandates reading the section).
7. create-newsletter “Voice discipline applies to every section including Brain Fodder” — restates the three filters and avoid-list examples. Keep the Brain-Fodder-is-not-a-looser-register point (skill-specific); trim the restated rules.
8. Light restatements of the three filters in write-communications-post and write-page-copy — borderline; one-line summaries with a pointer are acceptable, but trim the multi-line versions.

Nothing was found still saying em dashes are allowed “sparingly” — the one hit is create-email correctly describing the old rule as replaced.

Finding 4: Per-platform trajectory — no canonical home, and the closest candidate is stale

The known platform rules live scattered across memory (machine-local), the campaign guidelines (partly superseded), and social-media-tags.md. Status of each rule the prompt listed:

Rule	Where it lives now	Status
Instagram ~150w, “Link in bio”	Memory only (feedback-instagram-link-in-bio)	Not codified anywhere a fresh session loads. “Link in bio” convention, bio-link update requirement, no “on our site” closers: memory only.
Instagram hashtag placement	social-media-tags.md + campaign guidelines say first comment (2026-03-27); MEMORY.md index and the session prompt say in the post content	Direct conflict — needs Patrick’s ruling (Decision A).
#baseworks_method, never #baseworks	social-media-tags.md (canonical) + campaign guidelines	OK, codified
Facebook collaborations: two separate posts (Patrick personal first-person first, Baseworks page third-person later), URL included for link-preview	Memory only (feedback-collaboration-posts, revised 2026-04-23)	Not codified — and the campaign guidelines’ Publishing Sequence #4 states the superseded pattern (“collaboration posts between Patrick’s personal account and the Baseworks account… closing line in first person”). Also superseded for Instagram: no collab posts at all (Baseworks account only, unified voice).
Facebook/LinkedIn links carry no UTMs	Memory only (feedback-social-links-no-utm)	Not codified. Campaign guidelines’ UTM section is newsletter-scoped (correct) but never states the social exclusion.
LinkedIn personal-account article	Campaign guidelines Publishing Sequence #2	Partly codified but stale: says to set the “Originally published at” canonical field, which LinkedIn removed (memory, 2026-03-26; workaround is a provenance line at the bottom of the article). The ~120w feed post, SEO description ≤160 chars, and separate article deliverable file are memory-only.
Niche-building, not mass-appeal	Campaign guidelines “Content strategy” section	OK, codified
Newsletters: no Patrick/Asia sign-off	create-newsletter skill + campaign guidelines #10	OK, codified
Brain Fodder: varied closing prompts, no sensational language	Voice discipline: create-newsletter skill. Varied closing-prompt lead-ins: memory only — the campaign guidelines’ Brain Fodder structure (“Closing question”) never says to vary the label	Codify the vary-the-label rule in the guidelines
Group/participant posts: neutral register, no praise	session-summary skill 7e (thorough)	Codified in the skill; not in the unified guide (it was on the Finding-4 candidate list in the system audit but v2.8 didn’t include it — Decision D)
Never link practice.baseworks.com publicly	Memory only — and _blog-post-guidelines.md actively instructs the opposite: “The Practice Platform — when referenced in text (practice.baseworks.com)” under Internal links	Not codified + one doc contradicts it. Correct targets: baseworks.com/practice/ or the filming article.

Proposed canonical home: the “Platform-specific voice adjustments” + “Publishing sequence” sections of _campaign-creation-guidelines.md, rewritten with the current rules above and renamed so skills can point at one anchor (e.g., “Per-Platform Trajectories”). social-media-tags.md stays canonical for hashtags only. All skills and §2a pointers then reference those two.

Internal contradictions in the campaign guidelines to fix in the same pass:

• Body Sections table says Facebook ~200w; the platform table says ~400w. One number must win.
• The platform table’s LinkedIn row (~250w feed post) doesn’t match the current three-part article approach (~120w feed post + article + reshare).
• “Images are referenced via ![alt](url) — hosted externally (Google Drive, VPS FileBrowser)” — stale; the media pipeline is B2 → media.baseworks.com CDN with NAS originals.

Finding 5: Repetition scan — good within pieces, absent across sibling assets

create-campaign Phase 3.5 and create-newsletter Step 0 are solid within-piece/within-newsletter scans. But the unified guide v2.8 rule the prompt cites is cross-asset: recurring facts (the development-timeframe line, venue descriptions, method credibility lines) must vary in framing across the campaign’s sibling assets (newsletter section + 3 social posts + blog). No skill requires that check. Fix: add a “cross-asset recurring-fact variation” item to create-campaign Phase 3.5, referencing the unified guide entry and feedback-development-timeframe-variation’s rotation principle (which itself should be codified into the guidelines, since it’s memory-only).

write-blog-post has no repetition scan at all; it arrives via the Final Voice Pass step (Finding 1).

---

Part 2 — Per-skill notes (beyond the cross-cutting items)

• create-campaign — Strongest overall. Fixes needed: §2a rework (Finding 3.1), cross-asset scan (Finding 5), Final Voice Pass.
• create-newsletter — Template/build content is legitimately skill-specific and should stay. Fixes: WRITING-PROCESS load, trim Brain Fodder rule restatement, Final Voice Pass before Step 1 (HTML build), author-conditional personal guide.
• create-email — Two issues beyond Findings 1–3: (a) its Reference section names the 2026-03 STRIPO mock as “authoritative HTML reference,” while create-newsletter correctly says to use the 2026-04-23 build with all fixes applied, not the older export — stale cross-reference; (b) its convention “No signature line. Emails are administrative, not personal correspondence” sits against the unified guide’s email-signature convention (sign with whoever is sending) — possibly an intended distinction between reusable templates and personal correspondence, but it reads as a contradiction (Decision B).
• write-blog-post / write-communications-post / write-page-copy — Structurally compliant; need Final Voice Pass + duplication trims only.
• post-to-group — Add voice-guide loading; replace §1d table with a pointer. Also a non-voice bug found in passing: the Phase 2 confirmation-gate template says “Patrick (user_id=1) / Asia (user_id=8)” but the skill’s own Known Data table says Patrick=8, Asia=2 on practice.baseworks.com (user 1 is the baseworks.com site). The preview template should be corrected.
• draft-response — Trim duplicated conventions; add WRITING-PROCESS; a lightweight voice-pass report (inline, not Editorial-Notes-logged) fits its short-form output.
• create-event-post — Largest gap: no voice loading at all (Finding 2). Add pre-flight voice loading (WRITING-PROCESS + unified; personal guide rarely applies since event copy is unified-voice) and a Final Voice Pass over post_content, SEO title, meta description, and OG copy before Step 8 (publish).
• update-science-content — Well-designed (condescension scan is a model). Move the WRITING-PROCESS reference from Step 7 into pre-flight; add Final Voice Pass at Step 8.
• session-summary — Add voice loading (Finding 2); trim the em-dash recipe (Finding 3.4); add Final Voice Pass to Phase 5. Its group-post tone rules (7e) are thorough and can become the canonical statement if Decision D moves the register rule into the unified guide.

---

Part 3 — Companion doc findings

_campaign-creation-guidelines.md

Covered in Finding 4. Summary of edits: rewrite platform table + publishing sequence with current rules; fix FB word-count contradiction; fix LinkedIn canonical-field instruction; fix stale image-hosting line; add UTM social exclusion; add “Link in bio” / bio-link update; add practice.baseworks.com rule; add Brain Fodder closing-label variation; add development-timeframe variation principle; version-history entry.

_blog-post-guidelines.md

1. Stale spelling rule: line 70 says "practise" not "train" — superseded by unified guide v2.1 (2026-03-25): American English “practice/practicing” throughout. (Lines 179/181 are historical session logs; leave those.)
2. Practice Platform linking instruction violates current rule (Finding 4, last row) — replace with baseworks.com/practice/ / filming-article guidance.
3. The Open Tasks section carries a completed March session log for the Winter recap including pre-ban em-dash decisions (“remaining em dashes are all doing structural work”). Historical, but as an active checklist section it reads as current guidance. Propose moving the Winter-recap block to that article’s Editorial Notes or marking it superseded by the 2026-04-19 ban.
4. Needs a version-history row for whatever changes are approved.

communications-guide.md

1. Em dashes in live sample copy: the tone-escalation examples in §Assignment & Engagement use em dashes (“Just checking in—I noticed…”). These are templates people copy; they seed violations. The parallel examples in VOICE-GUIDE-PATRICK were already cleaned; sync these.
2. Banned metaphor in guidance: §Mid-Week Check-Ins instructs framing resources as “threads to pull on if you’re curious” — “threads to pull on” is a named example of a generated metaphor in the unified guide’s AI Generation Warning. Replace with a direct phrasing (“optional resources for anyone curious”).
3. “Lead with transformation, not logistics” (Core Principles) — “transformation” is Avoid-list register and the line defines by contrast. Rewrite positively (e.g., “Lead with what the program develops”).
4. Stale version reference: §Tone & Voice cites the unified guide as “(v2.5)” — drop the version number entirely so it can’t go stale again.
5. “The Teacher-Student Dynamic” heading and body use “teacher/student” — it quotes the published teaching principles, so this may be intentional source language (Decision C).
6. Version-history entry needed for approved changes.

social-media-tags.md

Hashtag content is fine and canonical. Two items: the Instagram placement line depends on Decision A; and if the campaign guidelines gain a per-platform trajectories section, add a cross-link both ways.

_archive-newsletter-guidelines.md

Clean. It governs archival of historical content (no drafting), so no voice-loading requirement applies. Note the related standing flag: archived issues are verbatim historical record — do not voice-correct them (consistent with the 2026-07-07 archive-import editorial-status convention). No changes proposed.

---

Part 4 — Proposed fixes (for approval)

Grouped so they can be approved wholesale or individually. Items marked ⚠ depend on a decision.

Group 1 — Final Voice Pass (Finding 1): add the pointer step to all 11 skills.

Group 2 — Voice loading (Finding 2): add missing loads to create-event-post, post-to-group, session-summary; add WRITING-PROCESS to create-newsletter, create-email, draft-response, update-science-content (pre-flight); make personal-guide loading author-conditional in create-email/create-newsletter; add “including inline chat drafts” wording to every loading step.

Group 3 — Canonical platform trajectories (Finding 4): rewrite the campaign guidelines’ platform section and publishing sequence with current rules; codify the memory-only rules (two-FB-posts, no-UTM-on-social, Link-in-bio, IG unified voice, practice-platform linking, timeframe variation, Brain Fodder closing labels); fix the three internal staleness items; point social-media-tags.md and the skills at it. ⚠ Instagram hashtag placement per Decision A.

Group 4 — De-duplication (Finding 3): rework create-campaign §2a into guideline pointers (and drop the dead feedback-facebook-no-linkaway reference); trim restated rules in post-to-group, create-email, session-summary, draft-response, write-blog-post, create-newsletter, write-communications-post, write-page-copy to pointers, keeping each skill’s process steps.

Group 5 — Cross-asset repetition scan (Finding 5): extend create-campaign Phase 3.5.

Group 6 — Companion doc corrections (Part 3): blog guidelines (practise line, practice-platform link, Open Tasks housekeeping); communications guide (em-dash samples, “threads to pull on,” “transformation” line, v2.5 reference ⚠ Decision C for the teacher section); create-email stale template reference ⚠ Decision B for the signature line.

Group 7 — Small bug: post-to-group confirmation-gate user IDs (Patrick=8, Asia=2).

Group 8 ⚠ (Decision D): codify group/participant-post neutral register into VOICE-GUIDE-UNIFIED as a v2.9 draft (pending Ksenia; requires Asia inbox item).

Process on apply: every skill touched gets both its table row and detailed section updated in 03-resources/claude-code-skills/index.md; version-history entries added where docs carry them; commit, push, changelog entry.

Decisions needed from Patrick

• A. Instagram hashtags: first comment (per social-media-tags.md + hashtag memory, 2026-03-27) or inside the post content (per MEMORY.md index + this audit’s session prompt)? Which is current?
• B. Email signatures in /create-email: is “no signature line” for reusable templates an intended exception to the sign-with-sender convention, or should the skill align with the unified guide? Resolved 2026-08-03 (Patrick): campaign/administrative/bulk emails are unsigned (Baseworks voice); personalized emails written in a personal voice are signed by the sender. Codified in the skill’s Conventions section.
• C. Communications guide “Teacher-Student Dynamic”: keep “teacher” as quoted source language from the teaching principles, or reframe the section around instructor/facilitator?
• D. Group-post neutral register: promote into VOICE-GUIDE-UNIFIED (v2.9 draft, pending Ksenia) now, or leave in the session-summary skill and memory?

Related

• Voice Guide System Audit 2026-07-07
• Content Skills Voice Audit — Session Prompt
• Unified Voice Guide
• Writing Process
• Patrick’s Voice Guide
• Campaign Creation Guidelines
• Social Media Tags Reference
• Communications Guide