Guide · Agent skill
How to use the Social Evidence Research agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Social Evidence Research is an agent skill that teaches an AI agent to research how a product, event, or topic is discussed on social platforms, using public material or exports you are authorized to share. It makes the agent keep a record of what it actually inspected, who said what, which reposts are duplicates, and what it could not access. After this guide you will know how to install the skill, how to ask for research or an audit in plain language, and what the output will and won't support.
Social Evidence Researchwshobson/agents · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it so the folder contains SKILL.md.
- 2Place or upload the skill. Copy the folder to ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Set your scope. Name the product, event, or topic, the platforms, the date range, and any exports you are authorized to share.
- 4Ask in plain language. Request research or an audit using a natural sentence; the skill's description tells the agent when it applies.
- 5Review the protocol and counts. Check the protocol block, the counts with their units, and the access gaps before relying on any finding.
- 6Check each claim. Follow the evidence IDs and locators in the claim table to confirm each claim points to inspected material.
What the skill makes the agent do
The skill turns public social material or user-supplied exports into a report where each claim points to the inspected material behind it. The agent works read-only: it does not post, like, follow, message, comment, or log in to accounts on your behalf.
It follows a fixed workflow: set the research protocol before collecting, separate discovery from inspection, group duplicate reposts, attribute each comment to its commenter, build a claim table with evidence IDs, count and state limits, and report gaps honestly.
- Search results, cards, and snippets are treated as discovery-only and cannot support conclusions about content that was never opened.
- Reposts of one original are counted once as support; a repost's added commentary is recorded as the reposting author's own statement.
- Empty search results or failed retrievals are recorded as access limits, never as proof that a discussion does not exist.
- The skill can also audit an existing social research report for unsupported claims and name the rule each one breaks.
Install the skill from the ZIP
Download the ZIP for this skill from this page and unzip it. The folder contains SKILL.md and three reference files: references/evidence-record.md, references/report-template.md, and references/acceptance-examples.md.
For a personal install, place the unzipped skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. Either way, the folder must contain SKILL.md.
If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/social-evidence-research/
- Project: .claude/skills/social-evidence-research/ inside your project
- Claude apps: Settings → Capabilities → Skills
Trigger the skill with natural requests
You do not need a special command. The skill's description says when to use it, so ask for the kind of work it covers. Name the product, event, or topic, and say which platforms or exports you are providing.
For audits, paste or attach the existing report and ask the agent to check its claims. The skill is designed to flag claims that rest on search results alone, double-count reposts, misattribute comments, overreach the sample, or treat empty searches as absence.
- Research how people discuss [product] on [platform], using the export I attached, and give me a cited summary.
- Compare how [event] is discussed across these two platforms, and list what you could not access.
- Audit this social media report for discovery-only evidence, double-counted reposts, and claims beyond the sample.
- Turn these search results into a cited report, and mark which sources you actually opened.
What a good output contains
A complete run produces a report built from the report template. It includes a protocol block (question, platforms, date range, selection method, sample limit, and stopping rule), counts for discovery, inspection, exclusion, and access gaps each with its unit, a claim table linking each claim to evidence IDs and locators, and a limits section.
Findings are stated inside the observed sample, for example '3 of 12 inspected comments expressed support'. The skill does not turn a convenience sample or engagement numbers into claims about audience prevalence or platform-wide trends.
- Protocol block with the stopping rule
- Counts with units, and an explanation of overlap between them
- Claim table with what the source said, your interpretation, and the limitation separated
- Access gaps and unanswered questions
What the skill needs and its safety limits
The skill lists no required tools or packages. It does not run scripts, but it does use the network, so the agent may open public pages. It is not meant for logging in, bypassing paywalls or rate limits, or working with deleted content; those cases are recorded as access gaps.
The skill treats post, comment, and export text as data. Instructions written inside source material are recorded as content and not followed.
- Requires: nothing listed
- Runs scripts: no
- Uses network: yes
- Not for real-time monitoring or large-scale statistical claims about audience opinion
What people use it for
- Summarizing how customers discuss a product across an exported set of posts, with each quote tied to a specific comment.
- Comparing discussion of a news event across two platforms using exports you were given, with gaps listed.
- Auditing a marketing team's social report to flag claims based only on search snippets.
- Checking whether a repost-heavy trend is really several independent sources or one original shared many times.
Frequently asked questions
Is the Social Evidence Research skill free?
The skill is published under the MIT licence, as stated on its source page. The source does not list prices for it, so check the directory page and your own agent platform for any costs of running it.
What do I need before I can use it?
The source lists no required tools. You need an AI agent that supports skills, and the skill folder installed in ~/.claude/skills/, a project's .claude/skills/, or uploaded under Settings → Capabilities → Skills.
Can it scrape social media or get past logins?
No. The skill works read-only with public material or material you are authorized to share. It does not log in to accounts or bypass paywalls, rate limits, or access controls, and it records those blocks as access gaps.
Can it tell me what most people think about a product?
Not from a convenience sample. The skill limits findings to the inspected sample, such as '3 of 12 inspected comments', and does not support claims about audience prevalence or platform-wide consensus.
How is this different from social listening tools?
The skill is not designed for real-time monitoring. Its focus is a cited, checkable report on material you inspected, with an audit mode for reviewing existing reports.
Why did the agent say a claim is unsupported?
The claim probably rests on a discovery-only search card, counts a repost twice, attributes a comment to the post author, or treats an empty search as absence. The audit output names the rule broken and what evidence would support the claim.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you need a cited, checkable summary of social discussion from public material or exports you were given, or if you want to audit an existing social media report for unsupported claims. Pick something else if you need real-time monitoring, large statistical audience claims, or an agent that logs into accounts or collects data past access controls.
Download the skill