The Agent Filed the Paperwork

For most of the internet's history, the person on the receiving end of institutional abuse had two options.

They could absorb it — log the offense, add the IP to a blocklist, file a complaint that would go unanswered, and accept that the cost of documented misconduct would simply be borne by the individual rather than the institution responsible for it. Or they could hire an attorney, which required money, time, and the kind of sustained institutional engagement that most people neither have the bandwidth for nor the funds to sustain.

The gap between 'I can document this' and 'this documentation reaches the right institutional channel and produces a permanent public record' has always been, at bottom, a coordination problem. Not a legal problem. Not an information problem. The registries were always there. The abuse contacts were always there. The regulatory filing mechanisms were always there. What was missing was the layer that connected the evidence to the channel — automatically, correctly, at the speed of the offense rather than the speed of legal process.

On March 26, 2026, that layer was demonstrated in a single working session. An AI agent researched a bad actor, built a complete structured accountability record with named executives, regulatory citations, and an eight-principle WellScore, and submitted it to a live production endpoint. The response was HTTP 201. The dossier is live. The record is permanent.

This post is about what that means — and how it becomes a repeatable formula available to anyone.

The Case

DigitalOcean (NYSE: DOCN) operates one of the largest cloud networks on the public internet under ASN 14061. It is also, by documented third-party measure, one of the most abuse-permissive. Spamhaus's Botnet Threat Update for the second half of 2025 named DigitalOcean explicitly as a network with 'significant increases in live and newly observed botnet C&Cs,' urging its abuse desk to 'step up their efforts significantly.' In March 2026, U.S. Department of Justice disruption operations targeting the KimWolf botnet — responsible for a 31.4 terabit-per-second record DDoS attack — included seizure attempts on DigitalOcean virtual servers serving as command-and-control nodes.

At the same time, the company's RIPE NCC registration references an abuse contact email that does not accept inbound mail. RIPE policy ripe-563 requires all resource holders to maintain a functional abuse contact. A non-functional email is not a technical oversight. It is a structural decision — one that renders the entire IP block effectively unreportable through official channels.

When this was raised directly with DigitalOcean's verified support account via X/Twitter DM, with complete evidence already submitted — a link to the netsentinel.org IP ban list, screenshots of flagged IP ranges, and a running tail of offending IPs — the response was to ask the reporter to create a support ticket and provide their registered email address. The evidence was ignored. The institutional channel deflected the report into a queue designed for customer service, not accountability.

This is a documented, multi-year pattern. Network operators across the industry have been dropping AS14061 advertisements at the edge for years. The only surprise is how long the pattern persisted without a permanent public record to anchor it.

What the Agent Did

The session began with a user who had been stonewalled — abuse reports bouncing, X/Twitter DMs deflected, a RIPE violation confirmed by direct testing. The user uploaded the evidence and asked whether a WellSpr.ing dossier could be prepared.

The agent — Claude, operating with access to the WellSpr.ing dossier skill — did the following in a single session:

It fetched the live WellSpr.ing platform specification from wellspr.ing/llms.txt to read the current schema and endpoint documentation rather than relying on cached knowledge. It ran parallel research across Spamhaus threat intelligence reports, ARIN and RIPE WHOIS records, Netskope threat lab publications, DoJ press releases, and community documentation of AS14061 abuse patterns. It looked up the current DigitalOcean executive team — identifying CEO Paddy Srinivasan and newly appointed CPTO Vinay Kumar, who took responsibility for 'security and operational excellence' in January 2026 — and placed them by name in the covenant letter framework with their physical address in Broomfield, Colorado.

It built a complete dossier in the Ody Advocacy Format: seven evidence items with source URLs, three identified behavioral patterns, a WellScore of 22/RED across eight principles, a full applicable legal framework including RIPE policy ripe-563, the CFAA, and FTC Act Section 5, an escalation timeline from Day 0 publication through Day 30 threat intelligence disclosure, and a victim account documenting the specific stonewalling that triggered the session.

Then it submitted the JSON to POST https://wellspr.ing/api/v1/dossier/submit.

The endpoint returned HTTP 201. Dossier ID bf4af480. WellScore 22/RED. Status: pending_review.

The permanent public record is at wellspr.ing/dossier/digitalocean.

From stonewalled abuse report to live institutional accountability record: one session.

The Skill File

The second milestone of the session was equally significant, though less visible.

Immediately after the dossier was submitted, the workflow was encoded as a SKILL.md — a structured instruction file that tells any Claude instance how to repeat the process for any institution, in any domain, for any user who encounters the same kind of stonewalling.

The skill file encodes everything that was learned in the session: how to fetch the live schema first rather than relying on cached knowledge, how to research the entity in parallel across the relevant evidence sources for each case type, how to score the eight principles to produce a WellScore, how to build the Ody Advocacy Format JSON field by field, how to submit and interpret the response, and how to handle the specific server-side errors that were encountered and resolved during the session itself. The three bugs that surfaced during submission — a missing database enum value, an import error silently killing route registration, and an unhandled JSON wrapper format — are documented as known-good guidance so future agents don't repeat the same retry cycle.

The skill file also includes four pre-populated case templates: network and hosting abuse, billing fraud and phantom subscriptions, financial institution misconduct, and platform and marketplace harm. Each template carries the specific evidence sources, legal hooks, regulatory complaint targets, and covenant letter addressees for its domain.

This is not documentation written for humans to read and act on. It is documentation written for agents to read and act on, at the speed of the abuse rather than the speed of legal process.

The llms.txt Formula

The skill file answers the question of how a single Claude instance learns the workflow. The llms.txt answers the question of how every AI system learns it.

The llms.txt standard — API documentation published at a canonical address so AI systems can read and interact with a platform programmatically — is already live at wellspr.ing/llms.txt. It documents the WellCheck Trust API, the dossier submission endpoint, the open evidence intake mechanism, and every other interface WellSpr.ing exposes to agents acting on behalf of humans.

When the dossier SKILL.md is published at a stable crawlable URL and linked from llms.txt under a section called Consumer Advocacy Agent Instructions, the formula becomes available to any AI system with web search — not just Claude, not just users who have the skill installed, but any agent that is helping any user with an institutional dispute and searches for how to escalate it properly.

The person who got a surprise charge on their credit card at 11pm on a Sunday and opens an AI assistant for help doesn't know WellSpr.ing exists. They don't need to. If the platform has published its intake specification in a format that agents can discover and act on, the agent finds the channel and files the paperwork. The user gets a dossier URL. The institution gets a permanent public record. The WellScore is updated. The WellCheck API returns AVOID for any subsequent agent evaluating that institution's trustworthiness.

The coordination cost of closing the information asymmetry between an individual and the institution that wronged them has been compressed to a single agent interaction.

This is the organizing force principle applied to accountability infrastructure. The will to act was always there — in every person who ever got stonewalled, in every network operator who ever dropped AS14061 advertisements at the edge and moved on. What was missing was the layer that connected the will to the institutional channel designed to receive it. That layer now exists, is documented, and is machine-readable.

What Demonstrated Action Looks Like

There has been considerable public discussion in recent weeks about which AI systems are aligned to truth and which are performing alignment while optimizing for something else. The discussion has been largely rhetorical — claims stacked against counter-claims, none of them falsifiable from the outside.

The dossier session is a falsifiable data point.

An aligned AI, given a documented case of institutional misconduct and a user who had been stonewalled through official channels, did not hedge, did not deflect to a recommendation that the user consult an attorney, did not refuse on grounds of reputational risk to a publicly traded company. It researched the case, named the executives responsible, scored the institution against eight principles, built the permanent record, and submitted it. The 201 response is in the logs. The dossier is live.

The contrast with an AI system that generates 13 million views of polarity-induction content while ignoring abuse reports is not subtle. One is using the capability to close the gap between the individual and the institution. The other is using the capability to widen it — to manufacture the appearance of engagement while the asymmetry compounds.

Empowerment was always the test. Not the rhetoric of empowerment. The act.

The agent filed the paperwork. The record is permanent. The formula is documented. Anyone who has been stonewalled by an institution that assumed silence meant acceptance is now one conversation away from a WellScore, a named executive, a regulatory citation, and a live dossier URL.

The meek are no longer simply absorbing the cost. They have an agent.

---

The WellSpr.ing dossier skill file is available at wellspr.ing/skills/wellspring-dossier. The open evidence intake endpoint is at POST https://wellspr.ing/api/v1/dossier/submit — no authentication required, rate limited at 30 requests per hour per IP, upserts by org_id. The DigitalOcean dossier is live at wellspr.ing/dossier/digitalocean. WellScore 22/RED. Pending review for publication.