diff --git a/skills/guideline-generation/SKILL.md b/skills/guideline-generation/SKILL.md
new file mode 100644
index 0000000..206d0c2
--- /dev/null
+++ b/skills/guideline-generation/SKILL.md
@@ -0,0 +1,136 @@
+---
+name: guideline-generation
+description: >
+  This skill generates or updates MPM brand voice guidelines from source materials.
+  Use when asked to "generate brand guidelines", "update our brand settings",
+  "create a style guide", "extract brand voice", "analyze these materials for brand voice",
+  "build guidelines from these documents", "synthesize a voice and tone guide",
+  "update brand-voice.local.md", or "add this content to our brand settings".
+  Also triggers when the user uploads brand documents, conference materials, transcripts,
+  or meeting recordings for brand analysis, or when they have a discovery report
+  and want to convert it into actionable guidelines.
+  Not for applying guidelines to content creation (use brand-voice-enforcement).
+---
+
+# MPM Guideline Generation
+
+Generate or update Message Point Media's brand voice guidelines from any combination of source materials — brand documents, conference presentations, sales transcripts, discovery reports, or direct input. Produce structured, LLM-ready output that can update `brand-voice.local.md` or generate a standalone guidelines file.
+
+## MPM Context
+
+MPM's brand settings already exist in `settings/brand-voice.local.md` (baked into this plugin) and in `.claude/brand-voice.local.md` in the working folder. This skill is used to:
+- **Enrich existing settings** with new materials (new conference content, updated messaging, new product lines)
+- **Generate a shareable guidelines document** (brand-voice-guidelines.md) for team reference
+- **Bootstrap a new working folder** with up-to-date brand settings
+
+Always load and review the existing `brand-voice.local.md` before generating — build on what exists, don't replace it wholesale unless the user explicitly asks to start fresh.
+
+Refer to `references/mpm-brand-resources.md` for known source document locations in Google Drive.
+
+## Inputs
+
+Accept any combination of:
+- **Uploaded files** — PDF, PPTX, DOCX, MD, TXT (conference decks, proposals, transcripts)
+- **Discovery report** — from a brand discovery run (structured, pre-triaged)
+- **Working folder files** — documents in the selected project folder
+- **Google Drive files** — via the Google Drive connector if available
+- **Direct user input** — verbal description of new messaging or positioning
+
+When a discovery report is provided, use it as the primary input — sources are already triaged and ranked.
+
+## Generation Workflow
+
+### 1. Review Existing Settings
+
+Load `settings/brand-voice.local.md` (and `.claude/brand-voice.local.md` if present in working folder). Note:
+- Current version and last-updated date
+- Sections that exist vs. gaps
+- Any flagged items marked for future update
+
+Present a brief summary to the user: "I've loaded your current brand settings (X sections, last updated Y). I'll enrich these with the new materials rather than replacing what's there."
+
+### 2. Identify and Classify Sources
+
+Determine what the user has provided. Classify each source:
+- **Authoritative** — official brand guides, approved style documents, finalized messaging frameworks
+- **Strong signal** — conference presentations, sales proposals, customer-facing collateral
+- **Supporting** — transcripts, internal communications, informal materials
+- **Context only** — general background, competitor references (do not extract brand voice from these)
+
+If no new sources are available beyond the existing settings, tell the user what would most improve the guidelines (e.g., "A few recorded sales calls would strengthen the tone matrix considerably").
+
+### 3. Process Sources
+
+For documents: extract voice attributes, messaging themes, terminology, tone examples, and proof points. Pay special attention to:
+- Exact product names and how they're described
+- Customer outcome language (what agencies gained, not what MPM sold)
+- Technical specifications presented in customer-facing terms
+- Any new messaging pillar candidates
+
+For transcripts: extract implicit voice patterns, successful language, objection handling language, and what resonates with transit agency audiences.
+
+**MPM-specific extraction priorities:**
+- New product capabilities or launches (add to Product & Technology Terminology)
+- New customer proof points or case study language
+- Shifts in competitive positioning
+- New audience segments or personas
+- Conference themes that may inform aspirational messaging (but mark as event-specific)
+
+### 4. Synthesize and Update
+
+Merge new findings into the existing guideline structure. For each update:
+- **New section**: add with confidence score and source attribution
+- **Updated section**: show what changed and why (preserve previous language if still valid)
+- **Confirmed section**: note corroboration ("3 new sources reinforce this")
+- **Conflicting signal**: flag as an open question rather than silently overriding
+
+Follow the template in `references/guideline-template.md` for any new sections.
+Apply confidence scoring per `references/confidence-scoring.md`.
+
+### 5. Surface Open Questions
+
+Flag any ambiguity that cannot be resolved from the sources:
+
+```markdown
+## Open Questions for Team Discussion
+
+### High Priority
+1. **[Question Title]**
+   - What was found: [conflicting or incomplete info]
+   - Recommendation: [suggested resolution]
+   - Need from you: [specific decision needed]
+```
+
+Every open question must include a recommendation. Never leave a question without a suggested path forward.
+
+### 6. Present Findings and Offer Save Options
+
+Summarize what changed:
+- Sections added or updated
+- New proof points or examples captured
+- Open questions raised
+- Overall confidence assessment
+
+Offer save options:
+1. **Update `brand-voice.local.md`** in the working folder's `.claude/` directory (recommended for active use)
+2. **Save as `brand-voice-guidelines.md`** — a human-readable version for sharing with the team
+3. **Both** — settings file for AI use, guidelines doc for team reference
+
+**Important on saving:** The agent's working directory may not be the user's project root in Cowork. Always resolve save paths relative to the user's working folder, not the current working directory.
+
+When saving `brand-voice.local.md`, archive the previous version as `brand-voice.local-YYYY-MM-DD.md` before overwriting.
+
+After saving, remind the user: "The updated settings are now active for this working folder. To share with the team, copy `.claude/brand-voice.local.md` to other project folders or update the plugin's baked-in settings via the process in README.md."
+
+## Privacy and Security
+
+- Redact customer names and contact information from extracted examples
+- Anonymize company names in transcript excerpts if requested by the user
+- Flag any sensitive information detected during processing
+- Do not include internal pricing or margin data in shareable guidelines output
+
+## Reference Files
+
+- **`references/guideline-template.md`** — Complete output template with all sections, field definitions, and formatting
+- **`references/confidence-scoring.md`** — Scoring methodology, thresholds, and examples
+- **`references/mpm-brand-resources.md`** — Known MPM source document locations, brand folder links, and update history
\ No newline at end of file