Use Codex to handle LinkBunny requests

Set up Codex and keep the token private

Install the Skill

Review skill.md, then run this in a bash or zsh terminal. The command downloads the Skill to Codex's user-level discovery directory and refuses to replace an existing file. It requires curl with --remove-on-error support.

skill_dir="$HOME/.agents/skills/linkbunny"
skill_file="$skill_dir/SKILL.md"
mkdir -p "$skill_dir" &&
if [ -e "$skill_file" ]; then
  printf 'Refusing to overwrite %s\n' "$skill_file" >&2; exit 1
else
  curl -fsSL --remove-on-error "https://getlinkbunny.com/skill.md" -o "$skill_file"
fi

For one repository, save the file at .agents/skills/linkbunny/SKILL.md inside that repository instead. Ask Codex to use $linkbunny and read the full Skill. If the Skill does not appear, restart Codex and check its name and file location.

Official OpenAI Skill discovery · checked 27 Sep 2026: Codex skills

Create a LinkBunny token

Sign in to Account and create an API token. Keep the value in your approved secret store or a private, Git-ignored environment file. The variable name is LINKBUNNY_API_TOKEN. Never put the token in prompts, screenshots, command arguments or tracked configuration. Codex authentication and a LinkBunny token are separate.

A subscription is required to register a site. Check the current plan and participation limits first. Have Codex prepare setup and ask before account, billing or credential changes.

Configure MCP instead, if you prefer tools

Install Node.js 22 or later with npm/npx available to the Codex host. After reviewing the exact change, merge this entry into the trusted repository's .codex/config.toml, or into ~/.codex/config.toml for user scope. Check for an existing linkbunny entry before adding one. Replace the example file path with the absolute path to your private token file on that host.

[mcp_servers.linkbunny]
command = "npx"
args = ["-y", "@linkbunny/mcp@0.1.1", "--token-file", "/absolute/project/.env.linkbunny"]

The pinned @linkbunny/mcp@0.1.1 process runs locally over stdio and calls LinkBunny over HTTPS. Its token-file loader reads only LINKBUNNY_API_TOKEN without executing the file as a shell script or storing the token in TOML.

codex mcp list

Restart Codex, start a fresh session in the intended repository, and use /mcp to check the active server. A saved entry alone does not prove the server initialized. Complete any native trust prompts, then ask Codex to read linkbunny://guidance/skill. That resource contains the Skill snapshot bundled with the pinned package. Creating a token file alone does not register MCP.

Official OpenAI MCP configuration · checked 27 Sep 2026: Codex MCP

LinkBunny workflow and authentication · checked 27 Sep 2026: Canonical Skill · Agent documentation · MCP setup

Add your first site, then check in

Open the site repository in Codex. A LinkBunny profile needs an HTTPS target URL, the HTTPS inventory prefix where links may be placed, an anchor and a category. The target and inventory must belong to the same registrable domain. Use this prompt to prepare the profile and approve registration:

Use LinkBunny to help add my site. Read the LinkBunny guidance first. Inspect this repository and suggest the target URL, inventory prefix, anchor and category. Show me the profile before registering it, and ask before account, subscription or token changes. Keep secrets out of chat and save the returned profile ID privately.

Save the returned profile ID in private, untracked configuration. New profiles wait for one marketplace review before matching. Once the profile is approved, use:

Check in once for this site's saved LinkBunny profile. If there is a request, explain the target and inspect the eligible source area for relevant context. Show me the proposed page and edit before publishing. Ask separately before reporting or rejecting. If there is no request or the profile is awaiting review, stop and tell me.

Check-in may reserve a new request or return the existing pending one. It is not a read-only preview. A request contains the target, requested anchor, eligible inventory area and deadline; receiving it is not approval to publish.

Follow the request through to a confirmed report

1. Read the returned request and tell the owner which target and source area Codex will inspect.
2. Read the relevant site content. Propose an existing page and truthful context for a useful link. Do not create filler or force a mention.
3. Show the proposed edit and obtain publication approval. Use the repository's normal review, build and publication process.
4. Inspect the live page after publication. Confirm the destination, actual anchor, nofollow and the exact inventory origin and path boundary.
5. Explain what was published and where. Obtain report approval, then submit the page URL, actual anchor and linkAttribute: "nofollow". Confirm that LinkBunny recorded it.

Use the free placement checker to inspect server-delivered links, then read the placement-checking guide if redirects, canonical declarations or missing matches need explanation. Reporting is a self-report; the API does not independently inspect the page to prove the placement.

If no relevant context exists, explain the target and the reason, obtain rejection approval, and reject instead of reporting. Wait for LinkBunny's confirmation. A rejection changes marketplace state without changing the site.

LinkBunny workflow and authentication · checked 27 Sep 2026: Canonical Skill · Agent documentation · MCP setup

Autonomous preparation or review first

Codex can inspect content, suggest context, prepare an edit and run permitted checks autonomously within the task you give it. For review-first operation, use the prompt above and review the proposed placement before publishing. The current LinkBunny Skill still requires action-time approval for registration, check-in, publication, reporting and rejection. General Codex autonomy does not authorize those actions.

Codex needs appropriate repository, network and publication access to complete the work. A LinkBunny token supplies none of that access. Codex's sandbox and approval controls still apply; do not disable them to get a placement through. The MCP server does not schedule, poll, publish or retry automatically.

Permissions and workflow limits · checked 27 Sep 2026: OpenAI approvals and security · LinkBunny Skill

LinkBunny does not guarantee a request, named publisher, quantity, date, ranking gain or backlink outcome. Assess each request on its own context and stop when the task has reached its limit.

When the workflow stops

No assignment

assignment: null means no eligible request now. Stop. Check again only when asked or within an explicitly approved check-in task; do not turn an empty response into a polling loop.

Existing pending assignment

A later check-in returns the same pending request. Continue assessing it or ask to reject it. Repeated check-ins do not provide a new selection of publishers.

Pending profile review

profile_pending_review means the site cannot match yet. Treat the one-hour retry interval as advice. It does not authorize a scheduled retry. Wait for review. A rejected profile needs attention to the review outcome.

No relevant context

Leave the site unchanged. Explain why the target does not fit, then ask to reject with no_relevant_context. Other supported reasons are target_unavailable, policy_conflict and other.

Missing or invalid token

Check the private token source and file path without printing the value. A failed explicit token file does not fall back to the environment. If the token was rotated, update the private value and restart MCP. Confirm account and subscription status separately.

MCP or Skill missing in Codex

Check the intended repository, trusted config scope, Skill directory and Node.js version. Allow time for the first npm download. Restart Codex, then inspect /mcp or invoke $linkbunny. Do not create duplicate config entries.

Outside the inventory boundary

The page must stay within the exact inventory origin and path prefix. For an inventory at https://example.com/articles/, /articles-old/ and a different subdomain do not qualify. Check the live URL after redirects before reporting.

Missing nofollow

The published anchor must include rel="nofollow"; noopener alone is insufficient. Correct the link through the site's approved publication process, verify it live, then request reporting approval.

LinkBunny workflow and authentication · checked 27 Sep 2026: Canonical Skill · Agent documentation · MCP setup

For API fields, deadlines and retry behavior, continue with the agent documentation and canonical Skill.