Install Sourcevine agent skills

Install free workflows for campaign reports, competitor content briefs and audience questions with Sourcevine MCP.

Choose a focused workflow for your assistant. All three skills use Sourcevine MCP for live data and keep source evidence in files you control.

Free to install. Sourcevine API usage consumes credits. Live workflows need a Sourcevine account, an API key, and an assistant with both local skill support and MCP tools. Skills run when requested; installation does not start a recurring schedule. You can try the fictional examples offline without an API key.

1. Connect Sourcevine

Create an API key, then follow the MCP setup guide. Connect once for all three skills. Keep the key in your client’s MCP configuration, never inside a skill or a prompt. Confirm your assistant can see the tools for your chosen workflow:

• Campaign Report: tiktok_stats, instagram_stats, youtube_stats, x_stats.
• Competitor Content Brief: tiktok_posts, instagram_posts.
• Audience Questions & Objections: tiktok_comments, instagram_comments, youtube_comments.

2. Install your skills

Each download is one readable Markdown file, with no executable installer. Install one or all three, each in its own folder:

For a command already filled in for your skill, open Campaign Report, Competitor Content Brief, or Audience Questions & Objections.

In the shared command below, set skill_name to an ID from the table. Run it once for each skill you want to install.

Codex

For a personal skill on macOS or Linux:

skill_name=sourcevine-campaign-report
skill_dir="$HOME/.agents/skills/$skill_name"
mkdir -p "$skill_dir" &&
curl -fL "https://sourcevine.io/skills/$skill_name/SKILL.md" \
  -o "$skill_dir/SKILL.md.download" &&
mv "$skill_dir/SKILL.md.download" "$skill_dir/SKILL.md"

See the official Codex skills guide for project scope and discovery settings.

Claude Code

For a personal skill on macOS or Linux:

skill_name=sourcevine-campaign-report
skill_dir="$HOME/.claude/skills/$skill_name"
mkdir -p "$skill_dir" &&
curl -fL "https://sourcevine.io/skills/$skill_name/SKILL.md" \
  -o "$skill_dir/SKILL.md.download" &&
mv "$skill_dir/SKILL.md.download" "$skill_dir/SKILL.md"

See the official Claude Code skills guide for project scope. These paths apply to Claude Code, not other Claude products.

On Windows, download the file manually into the same directory under your user home. Start a new client session after installation if the skill is not listed. Other clients need to support both the Agent Skills format and Sourcevine MCP; MCP support alone does not install the workflow.

These commands replace an existing skill only after the download finishes successfully. To update, run the same command again; save your own edits first if you have customized the file. To uninstall, remove just the chosen skill’s folder. Removing the skill does not remove your MCP connection or saved reports.

3. Run your workflow

In Codex, mention the skill ID prefixed with $; in Claude Code, use /. For example, $sourcevine-competitor-brief or /sourcevine-audience-insights. Each skill page has a starter prompt you can copy and customize.

The two research skills request up to 20 items per page. These are workflow budgets, not API prices or account-wide spending caps. Lower limits in your request take precedence. The assistant checks current prices before making calls and asks you to narrow a list that exceeds your scope. More pages or a larger scope require an explicit request and a new estimate.

For a campaign report, mention $sourcevine-campaign-report or /sourcevine-campaign-report and add your own public post URLs:

Create a campaign report for our October launch using these post URLs:
[paste the campaign post URLs here]

Use at most 10 credits.
Save the Markdown report and JSON snapshot.

The default limit is 10 unique posts and an estimated maximum of 20 credits. A lower limit in your prompt takes precedence. The assistant estimates the cost using current tool prices before making calls. It fetches each post once, preserves unavailable results, and stops on a tool error so you can resolve the cause. It may ask you to narrow a list that exceeds the limits.

All three workflows stop on tool errors and preserve partial results, marking the remaining inputs unattempted. A normal private or not-found result is recorded without retrying; it does not stop other authorized reads.

4. Compare campaign readings

Provide your earlier JSON snapshot together with the URLs:

Run the campaign report again for these same URLs.
Compare it with the attached previous snapshot.
Use a maximum of 10 credits.
Show changes only where the collection timestamps and metrics are comparable.

Each new reading consumes credits. Keep snapshots in your own workspace; installing the skill does not upload them to Sourcevine.

The skill checks the snapshot format before starting new API calls. Ambiguous duplicate posts, conflicting IDs or invalid timestamps are explained as comparison gaps. Platform changes use only matched, comparable posts; adding a new post to the list does not create artificial growth.

Try a comparison without API calls

Install the skill, then download the fictional previous snapshot and fictional current snapshot. Attach both to your assistant and ask:

Use the Sourcevine campaign report skill to compare these two fictional snapshots offline.
Treat campaign-report-baseline.json as the previous reading and
campaign-report-snapshot.json as the current reading.
Do not fetch any URLs or call MCP tools. Label the output as a fictional demo.

This demonstrates the reporting workflow without an API key or API credit usage; your assistant’s own usage charges may still apply. The expected view changes are TikTok +2,000, YouTube +500 and X −200. The Instagram post has no comparable reading. Baseline likes and comments were not supplied, so their changes stay unknown. Example URLs use example.invalid and are not real posts.

The JSON files include illustrative: true. Their recorded charges are fictional examples, not charges for running the offline comparison. Use snapshots from your own live reports for real campaign comparisons.

Try the research examples offline

Download the competitor evidence JSON or audience evidence JSON and attach it to your assistant with the matching skill installed:

Analyze this fictional evidence using the matching Sourcevine skill, offline.
Do not fetch URLs or call MCP tools. Label the output a fictional demo.
Explain the sample limits and link recommendations to evidence IDs.

The competitor example contains four posts and two proposed content tests. The audience example contains four usable comments, including two size questions and one cleaning concern. Neither is a real customer dataset or a representative market survey. An offline run makes no API calls; your assistant’s own usage charges may apply.

Get a decision-ready report

Each skill starts with a recommended next move and prioritized actions linked to evidence. Campaign Report gives you a follow-up queue and client update. Competitor Content Brief produces original hooks, outlines and test plans. Audience Questions & Objections drafts answers with placeholders for product facts you must confirm.

When your assistant can save files, the skills also produce a self-contained HTML report alongside Markdown and JSON (plus CSV for the competitor brief). Open the HTML in a browser, or use Print to save a PDF. It needs no external scripts or fonts. Ask for a specific format if you want fewer files.

Inspect a complete fictional report: Campaign Report, Competitor Content Brief, or Audience Questions & Objections. These demonstrate the output format, not real results. Suggested actions do not publish content or start scheduled runs.

Read the result

Each report includes post links, returned public metrics, collection timestamps, unavailable outcomes and reported charges. Missing fields remain unknown. Platform totals are separate, and summed views are never described as unique reach. A public post reading does not prove paid delivery or campaign impact.

View the illustrative report or download the example Markdown. The example is fictional and does not come from a customer campaign.

Not-found and private lookups are charged at the operation’s price. Retrieval failures are free. An interrupted tool call may not report its charge, so a report will mark that amount unknown rather than invent a total. See credits and billing and the API catalog for current prices.

Troubleshooting

• Skill not listed: check the folder name, SKILL.md capitalization and client discovery settings. A file downloaded to your Downloads folder is not installed.
• Tools unavailable: complete MCP setup in the same client where the skill is installed.
• Authentication or balance error: check your API key and account credits, then request another run. The skill does not retry automatically.
• No baseline: the first report establishes a snapshot. It cannot reconstruct earlier counts.
• Unsupported URL: supply canonical profile URLs for Competitor Content Brief, or canonical post URLs for the other skills. Ambiguous short links and unsupported platforms are skipped.

Last updated 5 October 2026