agentlinkops

A backlink ledger that lives in your repository. agentlinkops check fetches a source page, looks for the link you declared and records an observation with the HTTP status, redirect chain, robots posture and a SHA-256 of the fetched bytes. Page HTML never enters the repository. A check that cannot conclude says unknown with a named reason, and unknown never means the link was removed.

The CLI imports the same verifier module the hosted AgentLinkOps service runs, and its tests assert both paths return the same shape. The hosted checks, evidence store and MCP server are a separate service; this package is the local half and needs no account.

Give your agent AgentLinkOps

Use Node.js 22.13 or later. Choose a supplied-link check or a hosted task, then inspect the installed version's help before selecting guidance.

1. Inspect setup and choose guidance. CLI 0.6.10 installs the connection skill and its reference by default, with MCP setup for detected clients. Select optional guidance explicitly. Setup stores no credential and reports changes. CLI 0.6.9 and earlier can install the complete pack. Native-client acceptance remains separate from package checks.

   sh

   npx -y agentlinkops --help
   npx -y agentlinkops agent setup
   

   With the skills CLI instead: npx -y skills add https://agentlinkops.com, then add https://app.agentlinkops.com/mcp as a remote HTTP MCP server in your client. On a version whose help supports selection, use agentlinkops agent setup --skill ID or --recipe ID for optional guidance; repeat either flag for several choices. Choose --all-skills only for the complete pack. Customer instructions and records remain the customer's choice.

   Preview agentlinkops agent remove --recipe ID or --skill ID before adding --apply. The installer refuses skill removal while a recipe it owns still needs the skill. Keep the skill or select both that recipe and its skill after reconciling customer edits. The installer preserves edited recipe bodies, which can still block joint removal. For an intended complete uninstall, preview agentlinkops agent remove --scope project (or user), then add --apply.

2. Sign in for a chosen hosted task. The client opens the AgentLinkOps sign-in; you pick the account, workspace and scopes. Nothing is pasted into a chat. A supplied-link check needs no account or hosted connection.

3. Verify hosted access when needed. Ask your agent to list your AgentLinkOps projects, or run agentlinkops doctor.

4. Check a placement. Ask: "check whether [publisher page] links to [my page], show the dated observation and any uncertainty." No account yet? The same verifier runs locally:

   sh

   npx -y agentlinkops check --source https://publisher.example/resources --target https://your-site.example/guide --scope exact --json
   

   This fetches a public source page and returns an observation; no repository, ledger, CRM or publisher recipe is required. Read the state and evidence limits, including an unknown result. A returned result does not establish presence.

The skill is a short door: the first time your agent uses AgentLinkOps in a session it reads the full agent reference, which agentlinkops skill prints and https://agentlinkops.com/SKILL.md serves. Hosted monitoring is an invitation-only pilot; the local commands below need no account. The package is agentlinkops on npm, published by the agentlinkops account; the source is https://github.com/sunlover7/agentlinkops-cli.

Install the CLI on its own

npm install -g agentlinkops
agentlinkops --help

Or run it once without installing:

npx -y agentlinkops check --source https://publisher.example/resources --target https://your-site.example/guide --scope exact --json

Check one link with no account

Check a supplied pair from any working directory, without creating a ledger:

agentlinkops check --source https://publisher.example/resources --target https://your-site.example/guide --scope exact --json

If you choose to retain expected links in a repository ledger, use the optional local workflow:

cd your-repository
agentlinkops init
agentlinkops add --source https://publisher.example.com/resources --target https://example.com/guide --intent expected
agentlinkops check
agentlinkops status

init creates .agentlinkops/ with an empty ledger. add writes one entry. check fetches the source page and appends an observation to the mirror. status names every entry whose observations disagree with its intent and proposes the edit; applying it is a commit you make.

Ledger checks return 0 when no expected placement is conclusively absent and 1 when an expected placement is observed absent with complete evidence. Unknown observations do not fail a run by default. Pass --fail-on-unknown to return 1 for an unknown result in text or JSON output. Usage, configuration and malformed-ledger errors return 2.

agentlinkops doctor verifies the ledger, the verifier self-test, cloud reachability and the token, and prints a fix line for anything that fails.

The ledger

Intent and fact never share a file. You write intent; the verifier writes fact.

.agentlinkops/
  config.json          project identity, cloud origin, defaults. Yours, committed
  links.jsonl          the ledger: intent. Yours, committed
  observations.jsonl   fact mirror, tool-owned, committed
  events.jsonl         cloud event mirror by cursor, tool-owned, committed
  candidates.jsonl     imported and discovered rows, tool-owned, committed
  state.json           cursors and last-check marks, tool-owned, committed

One ledger entry:

{"id":"lk_9f3ac1","intent":"expected","source":"https://publisher.example.com/resources","target":"https://example.com/guide","scope":"exact","expect":{"anchor":"guide","rel":["nofollow"]},"cadence":"daily","ref":"q3-resource-pages/441","tags":["resource-page"],"added":"2026-09-11","note":"editor confirmed"}

Three intents: wanted (we want this publisher to link to us; present is the news), expected (it should be there now; absent is the alarm) and retired (stop checking, keep the history). Nothing promotes an entry automatically. id is random and permanent, so a corrected URL keeps its history. Everything is plain JSON Lines, one entry per line, that you can read without this tool.

Imports and adoption

Supplier exports import through presets or an explicit column map, and a preview writes nothing:

agentlinkops import ahrefs-export.csv --target example.com --from ahrefs
agentlinkops import any-export.csv --target example.com --map "source=Referring page URL,target=Target URL"

Presets: Ahrefs, Semrush, Majestic, Moz, DataForSEO, Linkody, Google Search Console and generic CSV. A preset is a convenience over --map, never a requirement. Every rejected row is reported with its line and reason. Count-per-site Search Console exports are refused, because importing them would invent placements.

An existing SQLite CRM (the reference adapter in scripts/agentlinkops.py) adopts into ledger entries read-only with agentlinkops adopt crm.sqlite; there is no reverse direction.

One page can be checked without a ledger or an account: agentlinkops check --source URL --target URL [--scope exact] [--json] [--out result.json] runs the same verifier and prints one result. --out reserves the file before the fetch and refuses to overwrite. A saved result later joins a ledger with agentlinkops adopt-result result.json [--intent wanted|expected], which validates the artifact (no raw HTML, a checksum that must match, a state consistent with its evidence) and appends the entry and observation once under a writer lock.

Sync with the hosted service

Local commands need no account. Sync pushes expectations and pulls events and evidence references into the mirror. It never writes into links.jsonl.

export AGENTLINKOPS_TOKEN=...   # from Agent access > Create API key in the hosted workspace
agentlinkops connect --workspace WORKSPACE_ID --project-id PROJECT_ID --origin https://app.agentlinkops.com
agentlinkops sync --dry-run
agentlinkops sync

Hosted access is an invitation-only pilot. The hosted checks, evidence store and MCP server are a separate service and are not part of this package.

Citation watches

An AI citation is a placement: a reference to your asset on a surface you do not control, re-decided by the model on every ask. So the CLI watches it the way it watches a link — a prompt panel, run against engine APIs on cadence, with dated evidence and honest statistics. A citation is never reported as "removed": each epoch states the citation rate with a Wilson interval and its sample size, classified against the previous epoch as declined, grown or not_distinguishable, or insufficient_data when n is too thin to interpret.

agentlinkops citation panel --domain example.com --brand Example --out panel.json
agentlinkops citation run panel.json --max-usd 1.00

New panels use the free mock engine. A live engine requires its own configuration and credentials; missing credentials never select mock results. The JSON panel contains targets, prompts, engines, optional samples and maxUsd. Answer snapshots live in .agentlinkops/citations/evidence/; observation and epoch ledgers live at the .agentlinkops/ root. The runner reserves estimated cost before each call and retry. A reported supplier overrun stops later calls; estimates do not enforce a provider invoice cap. A budget abort leaves a partial-epoch receipt and exit code 2. A cell that declined with complete evidence exits 1, like an expected link observed absent.

Browser measurement requires AGENTLINKOPS_PROXY_URL and refuses direct fallback. For an explicitly authorized direct diagnostic, set AGENTLINKOPS_BROWSER_EGRESS=direct-diagnostic and leave the proxy unset. The standard URL proxy provides no actual usage meter; its cost remains an estimate. Browser login and live account acceptance are separate.

Browser epochs retain supplied usage and cleanup receipts in .agentlinkops/citations/egress/<epochId>.jsonl. Unknown meter values stay null. A missing final record means the journal is incomplete; it does not establish complete costs or cleanup. Journals stop at 10,000 records or 8 MiB, and a failed write stops further samples.

Frozen inventories and evidence uploads check files against their retained observation hashes. Historical hashes verify the recorded answer fields and screenshot filename; they do not prove when an answer was captured or authenticate old screenshot pixels. Hosted uploads retain a separate hash of the exact uploaded file bytes.

Environment variables

No value is ever printed. The LINKTRAIL_* spellings, the linktrail command and a .linktrail/ ledger directory are still accepted with a one-line notice on stderr until 2026-10-15; run agentlinkops migrate once to rename the directory. It refuses to merge or delete anything.

Agent plugin

The package doubles as a plugin: .claude-plugin/plugin.json and .codex-plugin/plugin.json point at skills/, which holds the connect, CRM, campaign and asset skills plus the content-pipeline skills, with templates in templates/ and contracts in references/. agentlinkops agent setup installs the same skills for any detected client, and the discovery index at https://agentlinkops.com/.well-known/agent-skills/index.json serves them to npx skills add (the content pipeline sits behind https://agentlinkops.com/content). Node.js runs the ledger CLI; Python 3.10 or later runs the optional CRM helper and the watermark scanner.

What it does not do

It holds no backlink index, sends no email, verifies no contact address and runs no JavaScript: observations are static HTML only, and each one says so. Literal IP addresses, reserved hostnames and non-default ports are refused before any fetch.

Documentation and source

• Guides: https://docs.agentlinkops.com/guides/first-result and https://docs.agentlinkops.com/guides/connect-cli
• Product: https://agentlinkops.com/
• Source: https://github.com/sunlover7/agentlinkops-cli
• Contract references shipped in this package: references/cli.md, references/disavow.md, references/reports.md

License

Apache License 2.0. See LICENSE and NOTICE.