Define the report audience, decisions, and acceptance criteria
60-75 min
60-75 min
Decide who will use the recurring report and what decisions it should support before any code is written. Define the target category or company, competitor scope, run frequency, report owner, review deadline, historical comparison window, and approved distribution audience. Open Claude Code in Plan mode and paste the prompt below into the main terminal or chat input. Include the target URL, intended audience, required report sections, available hosting accounts, technical constraints, and examples of a useful competitive insight as source inputs. Ask Claude Code to propose the system boundary, data model, folder structure, execution sequence, failure modes, security model, and acceptance test. Copy the approved plan into `docs/ad-intel-plan.md` and save the Claude Code session reference in the project notes. Do not allow implementation to start until Marketing and the technical reviewer agree on what counts as a complete, trustworthy run.
An approved report and system specification with audience, decisions, scope, architecture, failure modes, security, and acceptance criteria.
Specify decisions, not just charts. A report that cannot change a campaign, positioning, budget, or research priority becomes an automated newsletter nobody needs.
Open Claude Code in Plan mode and create the implementation plan for a scheduled LinkedIn ad intelligence agent. Do not write or modify implementation code yet.
Source inputs:
- Target company or category URL: {{target_url}}
- Report audience and their decisions: {{report_audience_and_decisions}}
- Competitor scope and maximum count: {{competitor_scope}}
- Required report sections: {{required_sections}}
- Run frequency and historical comparison window: {{schedule_and_history}}
- Available accounts and platforms: {{available_platforms}}
- Technical constraints and preferred stack: {{technical_constraints}}
- Security, privacy, and data-retention requirements: {{security_requirements}}
- Examples of useful and useless insights: {{quality_examples}}
- Budget and operating-owner constraints: {{operating_constraints}}
Plan a system that uses the existing workflow tools: Claude Code, G2, TrustRadius, LinkedIn Ads Library, GitHub, Railway, Vercel, and Google Sheets.
Return:
1. System objective and non-goals
2. User stories and report decisions supported
3. End-to-end architecture
4. Repository and folder structure
5. Configuration and environment-variable design
6. Competitor discovery flow
7. LinkedIn collection flow
8. Raw-data schema
9. Snapshot and history model
10. Analysis taxonomy
11. Report data contract
12. Railway scheduling approach
13. Vercel report approach
14. Manual QA and Google Sheets archive
15. Logging, retry, partial-run, and alert design
16. Security and secrets handling
17. Local development and test strategy
18. Acceptance criteria for a trustworthy production run
19. Milestones in build order
20. Known risks, source limitations, and manual fallbacks
Rules:
- Keep observed ad evidence separate from strategic interpretation.
- Preserve raw snapshots before analysis.
- Design for partial collection and visible failures.
- Do not assume LinkedIn Ads Library structure or access will remain stable.
- Do not include credentials, cookies, or private data in the repository.
- Prefer a narrow, testable first production milestone over multi-channel scope.
Save the approved plan to `docs/ad-intel-plan.md` only after human review.