How to Combine Claude Cowork and a Client Brand-Guidelines Skill to Build a Monthly Recap
Use Claude Cowork to verify monthly client evidence, draft a source-grounded recap, then apply the client's own approved brand guidelines without changing facts.
Workflow summary
Quote this workflow accurately
- Best for
- Monthly client reports tend to fail in two predictable ways. A visually polished recap has too little proof of actual results, or a detailed source review takes so much work that the final message becomes inconsistent...
- Input
- A supported paid Claude Cowork plan: Pro, Max, Ultra, Team, or Enterprise, with the relevant features enabled. Skills have a separate eligibility policy and require code execution to be enabled; Free Skills access doe...
- Primary tool
- Claude Skills and Claude Cowork
- Output
- Use Claude Cowork to verify monthly client evidence, draft a source-grounded recap, then apply the client's own approved brand guidelines without changing facts.
- Main risk
- Return to the Cowork source map. Add actual deliverables or approved metrics; remove unsupported claims rather than dressing them up.
- Verification step
- Every metric has a source, unit, reporting period, and meaning checked by a person. Every screenshot supports the specific adjacent claim; it is not used as proof of unrelated outcomes. The factual draft has the full...
Continue from this article
Warning: A client recap can contain confidential revenue, performance metrics, client plans, screenshots, and unapproved claims. Scope access to one client and one reporting month, remove credentials and unrelated client information, and confirm organizational and client permission for cloud processing before providing materials. Do not send, publish, or promise client outcomes without review.
Cowork requirements at a glance: Claude Cowork on Pro, Max, Ultra, Team, or Enterprise (subject to account and administrator controls), an approved client workspace, actual source evidence, a private
monthly-client-recapskill or equivalent saved rules, optional client-specific brand rules, and a human reviewer. Access to a local folder requires Claude Desktop open and a desktop-started or scheduled Cowork session.
Problem and who this is for
Monthly client reports tend to fail in two predictable ways. A visually polished recap has too little proof of actual results, or a detailed source review takes so much work that the final message becomes inconsistent and late. Scattered invoices, screenshots, analytics exports, deliverables, and notes do not automatically tell a coherent story.
This workflow is for freelancers, consultants, and small agencies that report what was completed, the results supported by evidence, open blockers, next priorities, and decisions awaiting client input. The goal is a source-grounded monthly recap with repeatable structure and the right client's brand voice.
An important correction: which brand-guidelines skill?
Anthropic's public brand-guidelines skill is a real example, but it explicitly applies Anthropic's own colors, fonts, and visual identity. It is not a general-purpose way to style your client's report. Do not apply that public skill to an unrelated client's recap. For this article, "brand-guidelines skill" means a private client-brand skill based on the client's approved guide, or a brand-rules.txt file used as a manual style specification.
There are three layers. Cowork inspects authorized monthly evidence and produces a claim-to-source map. The private monthly-client-recap skill drafts repeatable sections from the reviewed evidence. A separate client-brand-guidelines skill, when desired and available, adjusts tone, names, and presentation after the factual recap is approved. Style cannot establish whether a claim is true.
Why Cowork fits this workflow
Cowork can inspect a carefully selected collection of deliverables, dated screenshots, metric exports, and project notes; separate evidence from claims; and identify missing support before drafting. Work can be organized across desktop, web, and mobile in modern Claude. In the unified experience, start by describing the task. If the account still offers a Cowork selector, select it.
New Pro, Max, and Ultra tasks run in the cloud by default. Files in a connected desktop folder remain on the computer until needed, and a cloud task can fetch the ones it needs through Claude Desktop when the session was started on desktop or is scheduled. That is not a promise that client data stays local. A mobile- or web-started session cannot directly enumerate an arbitrary desktop folder. Projects with local-folder ties support Cowork sessions on desktop, not every surface.
Why not normal chat or a single skill?
For a one-off recap, ordinary chat is sufficient if all needed evidence is already included. For recurring reporting, the difficult part is maintaining a narrow month-end source set, reconciling conflicting numbers, and proving where each client-facing statement came from. A writing skill cannot do that review merely by providing tone instructions. A brand skill cannot prove business results.
| Component | Job | Gate |
| --- | --- | --- |
| Cowork | Inspect permitted source files, inventory, prepare evidence summary and source map | Stop before drafting |
| Human reviewer | Check metrics, attribution, date ranges, and claim support | Approve factual evidence |
| Private monthly-client-recap skill | Draft sections and consistent client-safe wording | Only use approved evidence |
| Optional client-brand-guidelines skill | Apply the actual client's voice, naming, and presentation | Change style, not facts; recheck afterward |
| Human sender | Authorize and send the final recap | No automated external delivery |
Prerequisites
- A supported paid Claude Cowork plan: Pro, Max, Ultra, Team, or Enterprise, with the relevant features enabled. Skills have a separate eligibility policy and require code execution to be enabled; Free Skills access does not imply Free Cowork access.
- A single-client workspace and permission to process its contents in Anthropic's cloud. Check confidentiality agreements and organization policy before upload or desktop-file bridging.
- Current-month deliverables, dated metrics, screenshots with context, and approved notes. If a source is not permitted or unavailable, do not fabricate a replacement.
- A private
monthly-client-recapskill, or its saved instructions as a manual fallback. - Optional written brand specifications approved by that client. The public Anthropic brand-guidelines example is not an approved client brand guide.
- A person accountable for reviewing metrics, screenshots, claims, blockers, and commitments.
Set up the client workspace
Keep this working structure and use the same one for each reporting cycle:
source-material/
evidence-summary/
final-recaps/
archive-prior-months/
recap-source-notes.txt
brand-rules.txt
source-material/ contains only the current reporting month's approved deliverables, dated data exports, screenshots, and notes. Keep old months in archive-prior-months/, with archive content excluded from this month's facts unless a comparison is specifically authorized. evidence-summary/ holds the source map and unresolved claims. final-recaps/ stores approved final drafts. recap-source-notes.txt specifies the reporting period, approved metric definitions, date ranges, client-safe language, and known reporting limitations. brand-rules.txt contains only the client's approved style and naming rules.
Choose the access path deliberately:
- Desktop folder: Set up the client folder on your machine, connect only this folder, and start the task on Claude Desktop. Keep Desktop open during access. Review the task from mobile later if useful.
- Cloud-only session: Attach only the approved current-month files and rules documents directly, or use approved project knowledge; do not assume a connected local folder can be enumerated from a mobile-started session. Save downloaded results back to the client workspace manually when writeback isn't supported.
Confirm the specific files are available to the chosen Claude session, and that the client or organization permits Anthropic cloud processing and the retention settings in effect for this account. Do not treat a connected local folder as an offline processing guarantee.
The following JSON is a sample instruction block to paste into Claude, not a Claude API request or an installation file:
{
"project_name": "Monthly Client Recap",
"project_purpose": "Prepare a claim-to-source evidence summary and write a client-safe monthly recap only after evidence approval.",
"project_instructions": [
"Use only current-month source-material unless I approve an explicit historical comparison.",
"Create an evidence summary and source map before drafting any client-facing recap.",
"Separate measured outcomes, completed deliverables, qualitative observations, blockers, next priorities, and client decisions.",
"Include the source filename, reporting period, and basis for every number.",
"Treat screenshots as evidence only for the claim they actually depict.",
"Mark missing or contradictory information [CONFIRM] rather than inventing metrics, returns, or promises.",
"Wait for human approval before invoking the monthly-client-recap writing skill.",
"Apply the client's own brand rules only after the factual draft is approved.",
"Never send, delete, overwrite original source files, or promise outcomes.",
"Save reviewed drafts under final-recaps if this session can write to that folder."
],
"file_naming": "client-monthly-recap-client-name-yyyy-mm"
}
Build the private recap and brand skills
1. monthly-client-recap: the reporting-logic skill
Create a monthly-client-recap folder with a SKILL.md Markdown file. Anthropic documents name and description metadata at the start of a skill file, a ZIP with the skill folder at the root, and the ability to enable a skill under Customize > Skills. Use this starter:
---
name: monthly-client-recap
description: Draft a factual monthly client report from a human-reviewed source map with work completed, results, blockers, priorities and open decisions.
---
# Monthly Client Recap
Use after a reviewer approves the source map. Do not gather extra evidence or add unsupported claims.
## Required sections
- Work completed
- Outcomes or results supported by cited source data
- Open items or blockers
- Next priorities
- Decisions requiring client input
## Guardrails
- Do not invent metrics, cases, dates, guarantees or client results.
- Include reporting dates and keep estimates labeled as estimates.
- Keep screenshots tied to the specific claim they support.
- Mark unresolved or contradictory claims as [CONFIRM].
- Keep the tone client-safe, specific and non-promotional.
- Do not send the recap.
2. client-brand-guidelines: optional client-specific style skill
If approved brand rules are repeatedly needed, create a separate private skill for that client's actual brand, rather than importing Anthropic's public brand-guidelines example as the client's identity. Use an explicit client name in the final skill's metadata if you work with several clients. Keep confidential client branding assets out of other clients' sessions.
---
name: client-brand-guidelines
description: Apply only the supplied client's approved naming, tone, and layout rules to a previously fact-checked client recap, without altering its claims.
---
# Client Brand Guidelines
Use only after a human approves the factual recap. Read the approved client-specific brand rules attached to this task.
## Allowed changes
- Voice, tone, section hierarchy, spacing, labels, and approved product names.
- Style changes explicitly supported by the client's brand-rules.txt.
## Forbidden changes
- Do not change metrics, attribution, date ranges, promises, blockers, priorities, or open decisions.
- Do not borrow Anthropic's brand colors or typography unless the client has explicitly approved them.
- Do not add logos, images, case studies, or unsupported statements.
Return a revised draft for human review. Do not send it.
Review any downloaded skill code before enabling it. The public Anthropic skill-creator is an example tool for drafting skill packages, not a required dependency. Anthropic's custom skills guide covers ZIP packaging and activation. Skill triggering is selected by Claude, so naming the skill in a prompt does not guarantee it ran.
Confirm the account's current custom skill upload/enable path, accepted skill file capitalization, and whether both private skills are available and actually invoked on the planned surface. If not, use their written instructions directly as two distinct prompt stages.
How to gather source evidence
- Limit the workspace to one client and one reporting period. Remove credentials, internal margin calculations, and unrelated client files.
- Include only data exports with dates and defined units. Distinguish a screenshot of an analytics dashboard from an independently verified outcome.
- Name deliverables and screenshots so the reported work can be traced to an actual file; note date ranges, reporting systems, and version numbers.
- Use
recap-source-notes.txtto identify which metric source is authoritative when exports disagree, and list client-approved terms and promises. - Keep
brand-rules.txtseparate from the evidence, so styling instructions do not contaminate factual reasoning. - Keep old files in
archive-prior-months/or outside the accessible project, not mixed with this month's results.
Numbered workflow: source review first, brand polish last
1. Ask Cowork to inspect only the current reporting month
Use a desktop-started task for a connected local folder, or a cloud session with files already supplied. Request an evidence summary and map, not a finished email:
{
"task": "Inventory the client's current-month material and prepare evidence for a monthly recap",
"input": {
"source_set": "source-material or explicitly attached current-month files",
"reporting_period": "YYYY-MM-01 through YYYY-MM-last-day",
"source_rules": "recap-source-notes.txt",
"excluded_material": "archive-prior-months and unrelated clients"
},
"rules": [
"Identify all source filenames and available date ranges before making claims.",
"Separate delivered work, measured outcomes, qualitative observations, blockers, and open decisions.",
"Record the source filename for each claim and numeric metric.",
"For every metric, note its unit, reporting period, and numerator/denominator if applicable and supported.",
"Identify conflicting exports and ambiguous screenshots as [CONFIRM].",
"Flag missing support; never invent a metric, client result, case study, or deadline.",
"Do not write the final recap or change source files."
],
"output": {
"source_inventory": "File and reporting-period list",
"evidence_summary": "Supported facts, deliverables, metrics, blockers, and open decisions",
"source_map": "Claim to filename, reporting period, and supporting location",
"unresolved_items": "Missing, weak, or conflicting support for human decision"
}
}
If Claude cannot inspect the files, report blocked, not a recap based on last month's report or generic client language. Where possible save the source map in evidence-summary/, or download it and save manually.
2. Have a person verify the evidence
Open the underlying exports and check exact values, units, time windows, attribution, and dates. Confirm whether figures describe output (for example, emails sent) or real results (for example, verified sales). Do not convert correlation into causation or a screenshot into proof of a business outcome. Resolve contradictory numbers or mark them [CONFIRM]. Remove sensitive context the client does not need to see. Approve the factual summary explicitly.
3. Test the recurring recap skill with sanitized evidence
Give monthly-client-recap a short approved example containing one deliverable, one confirmed metric, one blocker, and one decision. Check that it preserves the numbers and includes all required sections. Revise its description or instructions if it is not invoked. This test confirms behavior for the current setup; it does not prove every future run will invoke the skill.
4. Draft from only the approved evidence
{
"task": "Use the private monthly-client-recap skill to draft a client-facing report from approved evidence",
"input": {
"approved_evidence": "PASTE OR ATTACH HUMAN-APPROVED EVIDENCE SUMMARY AND SOURCE MAP",
"period": "REPORTING MONTH",
"recap_skill": "monthly-client-recap, if enabled"
},
"rules": [
"Use approved evidence only; do not add unsupported metrics, claims or promises.",
"Include Work completed, Outcomes or results, Open items or blockers, Next priorities, and Client decisions.",
"Keep dates, source references, and limitations in the draft or review notes.",
"Leave uncertain claims [CONFIRM] and never turn them into promotional certainty.",
"Do not apply brand polish yet.",
"Do not send or publish the report."
],
"output": {
"factual_recap_draft": "Client-safe recap structure with sourced claims",
"review_flags": "Remaining uncertainties and sensitive wording"
}
}
5. Approve the factual draft, then apply client brand rules
Recheck the factual draft. After approval, use the private client-specific brand skill if it is available, or apply brand-rules.txt manually. This third step is deliberately separate from drafting so that tone and style cannot conceal unsupported results.
{
"task": "Apply the client's approved brand rules to the reviewed monthly recap without changing its facts",
"input": {
"approved_factual_recap": "PASTE OR ATTACH APPROVED FACTUAL RECAP",
"brand_rules": "client-specific brand-rules.txt",
"brand_skill": "private client-brand-guidelines, if enabled"
},
"rules": [
"Change tone, headings, naming and layout only when the client approved the specific rule.",
"Preserve every numeric metric, reporting period, claim, blocker, priority and open decision.",
"Do not use Anthropic's public brand-guidelines skill as client branding.",
"Do not add case studies, images, logos, guarantees or results that were not supplied.",
"Return a revised draft and a list of substantive changes, if any, for review.",
"Do not send the recap."
],
"output": {
"branded_recap_draft": "Fact-preserving client-branded recap",
"change_review": "Any content that changed beyond style, requiring correction"
}
}
6. Perform the final source and client-safety check
Compare the branded recap to the approved factual version, with special attention to numbers, attribution, named products, client commitments, and omitted blockers. Reopen the source map for contentious claims. If brand editing introduced any factual difference, undo it or obtain a new approval. Confirm the audience should receive the included metrics and screenshots.
7. Save the finished recap; let a human send it
Use client-monthly-recap-client-name-yyyy-mm. Save the approved final draft, factual draft, and source map in the client reporting workspace when supported; otherwise download them and save manually. A human sends the recap through the authorized client channel. Archive old evidence deliberately and do not let the model delete the source records.
Make monthly preparation repeat without automating the approval gates
A recurring schedule is helpful when it prepares an inventory and source map. In Claude use the Scheduled sidebar to create or manage a monthly task. Anthropic removed /schedule from the slash menu on October 29, 2026; entering it can fill in a scheduling request, but the Scheduled sidebar is the stable primary instruction.
- Cloud-based monthly task: Can run without your computer awake if the required sources are already accessible in the cloud and the task has appropriate permissions.
- Connected local client folder: A scheduled task still needs Claude Desktop running for local file access. If the desktop bridge is unavailable when the run starts, the task must flag missing access and stop rather than reuse historical claims.
- Mandatory human handoffs: Use the scheduled run to prepare evidence only. After a person verifies the source map, run the recap-writing skill. After another review, run optional client brand polishing. Finally, verify and send manually.
Confirm the Scheduled feature is enabled on the actual account, that the scheduled run can read its approved source set, that required skills are available in that session, and that client contracts and organization rules permit that processing. If not, use a manual recurring task.
For future months reuse this preparation prompt:
{
"task": "Prepare the current month's client recap evidence for human approval",
"scope": "Single client, current-month source-material only",
"rules": [
"Inventory sources with dates, metrics, units and claim-to-source references.",
"Separate verified outcomes from observations, blockers and open decisions.",
"Flag conflicts, missing evidence and anything uncertain as [CONFIRM].",
"Do not draft the client-facing report, apply brand polish, or send anything.",
"If client sources are inaccessible, state blocked and request access correction."
],
"deliverables": [
"Evidence summary",
"Source map",
"Unsupported-claims list"
]
}
Expected output
Each reporting cycle should produce the current month's evidence summary, claim-to-source map, unresolved claims list, a factual recap, a separately reviewed client-branded version if requested, and a saved review record. The final report consistently includes Work completed, Outcomes or results, Open items or blockers, Next priorities, and Decisions requiring client input. No claims are added merely because a brand template expects them.
Tool-specific instructions, permissions, and safety
Use Manually approve for this sensitive client workflow. The other current Cowork options are Automatically approve (uses safety checks and may ask) and Skip all approvals (no automatic checks). Automatically approve is not equivalent to skipping approvals. File deletion still requires explicit permission. The human factual and final review stages remain mandatory regardless of mode.
Claude's plugins are saved to the account, but no plugin is required by this article. If you connect an external dashboard, CRM, or file source, review its permissions separately. Plugins are not a justification to gather unrelated client material. For broader context, see the Cowork tool hub, Claude Skills hub, and Anthropic Workflows series.
Quality checks
- Every metric has a source, unit, reporting period, and meaning checked by a person.
- Every screenshot supports the specific adjacent claim; it is not used as proof of unrelated outcomes.
- The factual draft has the full recurring section structure and no unsupported wins, dates, guarantees, or pricing.
- The brand layer uses the actual client's approved rules, not Anthropic's public color and font template.
- The branded draft contains the same claims, metrics, open decisions, blockers, and priorities as the approved factual draft.
- Archived months and unrelated clients are excluded.
- No external message is sent automatically, and confidential information is authorized for its intended audience.
- A person, not the model, decides which contradictory metrics are authoritative.
Common failure modes and fixes
The recap is stylish but evidence-light
Return to the Cowork source map. Add actual deliverables or approved metrics; remove unsupported claims rather than dressing them up.
The client receives Anthropic's branding
Do not use the public brand-guidelines skill as a generic client style tool. Replace it with the client's approved brand-rules.txt or private client-brand-guidelines skill.
Two exports report different values
Compare periods, metric definitions, filters, and units. Have the owner identify the authoritative source before the factual recap is approved.
A screenshot is mistaken for proof of results
Document what the screenshot actually shows and its date. If the claimed outcome is not supported, mark it [CONFIRM] and remove it from the client version.
A scheduled run cannot read the client's month-end files
Check whether the source is cloud-available or whether Claude Desktop is open for the local-folder bridge. Report blocked and run again after access is restored. Never recycle prior-month evidence as though it were current.
Skills do not appear to activate
Confirm they are enabled under the account, test sanitized inputs, and check whether the description matches the request. Use the same instructions manually if the surface does not activate the skill.
Sources Checked
- Anthropic, Get started with Claude Cowork (accessed 2026-11-01): https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork
- Anthropic, Use Claude Cowork on web, desktop, and mobile (accessed 2026-11-01): https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile
- Anthropic, Claude Cowork product guide: Getting started (accessed 2026-11-01): https://claude.com/resources/guides/claude-cowork-product-guide/getting-started
- Anthropic, Cowork changelog (accessed 2026-11-01): https://claude.com/docs/cowork/changelog
- Anthropic, What are skills? (accessed 2026-11-01): https://support.claude.com/en/articles/12512176-what-are-skills
- Anthropic, How to create custom skills (accessed 2026-11-01): https://support.claude.com/en/articles/12512198-how-to-create-custom-skills
- Anthropic, Use plugins in Claude (accessed 2026-11-01): https://support.claude.com/en/articles/13837440-use-plugins-in-claude-cowork
- Anthropic public
brand-guidelinesskill (Anthropic identity, not client branding; accessed 2026-11-01): https://github.com/anthropics/skills/blob/main/skills/brand-guidelines/SKILL.md - Anthropic public
skill-creatorexample (accessed 2026-11-01): https://github.com/anthropics/skills/blob/main/skills/skill-creator/SKILL.md
Quarterly Refresh Flag
Review on 2027-02-01 for current Cowork file handling, scheduled tasks, private skill setup and triggering, retention and confidentiality guidance, and any changes to Anthropic's public brand-guidelines example.
Related Workflows
How to Turn Approved Feedback Into a Working Style Guide With Claude
Turn approved client feedback, example references, and call notes into a short working style guide that keeps execution consistent across future drafts.
How to Turn Analytics Dashboard Screenshots Into a Plain-English Client Value Summary With ChatGPT
Turn analytics dashboard screenshots into a plain-English client value summary that explains what changed, why it matters, and what to do next.
How to Turn a Client KPI Export Into a Quarterly Business Review Deck With NotebookLM
Use NotebookLM to turn a client KPI export, win log, and goals note into a source-grounded quarterly business review narrative and deck.
Help keep PromptedWork sharp
Share a broken step, outdated prompt, or general feedback. This is only for improving this specific workflow.