Ctxd

ctxd

Use ctxd first when the task is to extract context from supported work URLs.

Supported URL patterns:

• https://github.com/<owner>/<repo>/pull/<n> — GitHub PR
• https://*.slack.com/archives/... or https://*.slack.com/client/.../thread/... — Slack thread
• https://*.atlassian.net/wiki/... — Confluence page
• https://*.atlassian.net/browse/<KEY> — Jira issue

GitHub support is pull requests only. Repo, file, blob, issue, gist, and commit URLs are not supported — ctxd exits with Unsupported URL. To read a public repo file or README, fetch the raw URL with curl / WebFetch instead; do not route it through ctxd.

Why use it

• It exports dense Markdown or text instead of forcing the model through many tool calls.
• It leaves a stable local artifact that can be re-read, diffed, or shared.
• It is better than connectors for bulk context export.

Required config

Do not assume ctxd works without credentials.

Config file:

• ~/.config/ctxd/config

Slack:

• SLACK_TOKEN

Confluence and Jira:

• CONFLUENCE_BASE_URL
• CONFLUENCE_EMAIL
• CONFLUENCE_API_TOKEN

GitHub PR:

• GITHUB_TOKEN (classic PAT, repo scope)

If auth is missing, say which key is required. ctxd does not use the gh CLI, so gh auth login / gh auth switch has no effect on it, and the current working directory never changes which account it fetches as.

Default commands

Single item to stdout (cross-source recursion is off by default — only the primary URL is fetched):

ctxd '<url>' -f text

Do not add -q: when stderr is not a TTY — which is the case whenever an agent captures the output — ctxd silences progress logs on its own. Warnings and the completeness summary still print, and you should read them.

Enable recursion to auto-expand supported URLs found in the output:

ctxd '<url>' -f text --recurse-depth 1

Confluence page tree:

ctxd '<confluence-url>' -f text -r -O

Jira issue with its attachments (images, diagrams, PDFs) saved locally and the body links rewritten to point at them:

ctxd '<jira-url>' --all-attachments -O

Attachments are never downloaded by default. When an issue or page has attachments that were skipped, the run summary says so — read it before concluding that content is missing.

Profile a slow export:

ctxd '<url>' --profile

When to prefer ctxd over connectors

Prefer ctxd when:

• the user wants a lot of content at once
• the output should be saved to disk
• the user wants a recursive Confluence export
• the fetch should be reproducible outside the current chat

Prefer connectors when:

• the user wants one small field
• the user wants an in-product write action
• the task is interactive navigation rather than export

If ctxd fails or the URL is not a supported source, say what failed in one line and fall back to WebFetch / a connector — do not ask the user to paste the content by hand.