Competitive Analysis

Use When

• The competitor analysis tables already exist.
• You need to poll competitive_crawl_trigger on a schedule.
• You need standardized reports grouped by category.
• You need to send summaries to a DingTalk robot.

Do not use this skill for:

• one-off ad hoc analysis
• open-ended research without database inputs
• flexible report generation without a fixed template

Required Inputs

• Database connection: COMPETITIVE_ANALYSIS_DSN
• OSS endpoint: OSS_ENDPOINT
• OSS bucket: OSS_BUCKET
• OSS access key id: OSS_ACCESS_KEY_ID
• OSS access key secret: OSS_ACCESS_KEY_SECRET
• DingTalk webhook: DINGTALK_WEBHOOK
• Optional DingTalk signing secret: DINGTALK_SECRET
• In OpenClaw, prefer environment injection through skills.entries.competitive_analysis.env

Goal

1. Find unconsumed trigger rows where status='success'.
2. Load the last 6 months of product, review, and QA data.
3. Aggregate results by category.
4. Produce analysis_context.json for the host to continue narrative generation.
5. Generate a Markdown/HTML skeleton that follows the reference PDF structure.
6. Send a DingTalk summary.
7. Mark trigger rows as consumed after success.

Entry Points

Primary command:

python3 {baseDir}/scripts/run_report.py

Common arguments:

• --category CATEGORY
• --since-months 6
• --limit 20

Files

• SKILL.md: skill entry instructions
• references/report-outline.md: report structure contract
• references/data-contract.md: data contract and field expectations
• references/openclaw-setup.md: OpenClaw setup example
• scripts/run_report.py: main CLI
• scripts/render_report.py: Markdown/HTML rendering
• scripts/send_dingtalk.py: DingTalk delivery
• analysis_context.json: structured analysis context for the host runtime

Rules

• Follow the reference PDF for section order.
• If fields are missing, keep the section and mark values as 未采集 or 待补充.
• Keep the CLI stateless and let an external scheduler trigger it.
• Do not call any external LLM API from the script.
• Let the host runtime generate deeper narrative content from analysis_context.json and references/report-outline.md.
• In OpenClaw, prefer host-managed environment injection over .env.

Minimal Workflow

1. Read references/data-contract.md.
2. Confirm that the trigger table already includes the consumption fields.
3. Configure skills.entries.competitive_analysis.env as shown in references/openclaw-setup.md.
4. Start a new OpenClaw session so the skill reloads.
5. Run python3 {baseDir}/scripts/run_report.py or invoke it from an external scheduler.
6. Read the generated analysis_context.json.
7. Let the host runtime generate the final narrative based on references/report-outline.md.
8. Validate the final output against the report outline.

Success Criteria

• New successful trigger rows are detected.
• Reports are generated per category.
• Section structure matches the reference report.
• DingTalk receives the summary message.
• Trigger rows are marked as consumed.