Skip to content
BranchesTags

Repository files navigation

ctxd

Turn work URLs into LLM-ready Markdown.

📖 中文文档

ctxd overview

GitHub PRs, Slack threads, Confluence pages, and Jira issues — one command, one Markdown file you can save, diff, or feed to a model.

Agent skill (recommended)

ctxd ships a companion skill at skills/ctxd/SKILL.md that works with both Claude Code and Codex CLI. It teaches the agent to reach for ctxd whenever a supported URL appears in the conversation, instead of falling back to chat-style fetching or in-model connectors.

# Claude Code
mkdir -p ~/.claude/skills && ln -s "$(realpath skills/ctxd)" ~/.claude/skills/ctxd

# Codex CLI
mkdir -p ~/.codex/skills && ln -s "$(realpath skills/ctxd)" ~/.codex/skills/ctxd

The skill assumes Required config is already set up — if credentials are missing, the agent will tell you which key it needs before attempting a fetch.

Why ctxd

  • CLI-first, not chat-first — one command produces a stable Markdown artifact you can inspect, diff, archive, or feed into any model. No drip-feeding through many tool calls.
  • Comments and metadata stay attached — PR reviews and inline threads, Slack threading, Confluence attachments and page metadata, Jira custom fields all preserved.
  • Bulk export is the default — page trees, long threads, heavy PRs come out as one file. Parallel fetch across all sources (see Performance for numbers).
  • Never silent on data loss — fetch failures, skipped items, and truncations always warn on stderr and show up in the run summary / manifest.json, even under -q.

When CLI beats connectors

Situation ctxd CLI In-model connector
Export a whole Confluence page tree Best fit Usually many tool calls
Dump a long Slack thread for later summarization Best fit Usually repeated fetch + resolution
Save a PR review artifact to disk Best fit Usually no persistent artifact

Quick examples

# GitHub PR -> markdown file
ctxd -O https://github.com/owner/repo/pull/123

# Slack thread -> stdout
ctxd https://app.slack.com/client/T.../C.../thread/C...-1234567890.123456

# Confluence page tree with images -> local directory
ctxd https://your-site.atlassian.net/wiki/spaces/SPACE/pages/123456 -r -i -O

# Jira issue with referenced images -> local file
ctxd https://your-site.atlassian.net/browse/PROJECT-123 -i -O

Supported Sources

Source URL Pattern
GitHub PR https://github.com/<owner>/<repo>/pull/<number>
Slack Thread https://*.slack.com/archives/... or .../client/.../thread/...
Confluence https://*.atlassian.net/wiki/...
Jira https://*.atlassian.net/browse/<KEY>

Installation

Homebrew (recommended)

brew tap cheerchen/tap
brew install ctxd

From source

cd ctxd
uv sync --group dev

Required config

Before an agent can actually fetch anything, the relevant auth must already exist.

Config file:

~/.config/ctxd/config

Typical keys:

SLACK_TOKEN=xoxp-...
GITHUB_TOKEN=ghp_...
CONFLUENCE_BASE_URL=https://your-site.atlassian.net
CONFLUENCE_EMAIL=you@example.com
CONFLUENCE_API_TOKEN=your-token

Every key can also come from the environment, which takes precedence over the file. chmod 600 the file — ctxd warns on stderr if it is group/world readable.

Global Options

Option Description
-o, --output <path> Write to file/directory (default: stdout)
-O, --auto-output Auto-generate output path by source (mutually exclusive with -o)
-f, --format text|md Output format (default: md)
-q, --quiet Suppress progress logs only — warnings and the completeness summary always print (auto-enabled when stderr is not a TTY)
-v, --verbose Verbose logging
--profile Print stage / HTTP timing summary
--max-concurrency <N> Cap parallel work across fetchers (default: 5)
--recurse-depth <N> Cross-source recursion: expand supported URLs found in output (default: 0=off, max 2; opt-in with 1/2)
--no-recurse Disable cross-source recursion (equivalent to --recurse-depth 0; kept for explicitness)
--max-chars <N> Cap output characters (default: 100000 for stdout; file output unlimited unless set explicitly; -1 = unlimited)
--max-file-size <N> Cap per-attachment download size in bytes (default: 52428800 = 50 MiB; -1 = unlimited)
--max-run-size <N> Cap total attachment download size per run in bytes (default: 524288000 = 500 MiB; -1 = unlimited)

Options can be placed before or after the URL (e.g. both ctxd -q <url> and ctxd <url> -q).

GitHub PR

Export PR metadata, reviews, inline comments, timeline comments, and code changes.

Prerequisites

A GitHub token in GITHUB_TOKEN (env or ~/.config/ctxd/config). Create a classic PAT with the repo scope:

GITHUB_TOKEN=ghp_...

repo covers every repository you have access to plus all public ones. For a repository in a SAML-protected organization, also authorize the token for that org from the token list page, or the API returns 404 as if the repo did not exist.

Avoid fine-grained PATs here: they are limited to a single resource owner, so a token scoped to one organization cannot read public repos elsewhere.

Usage

ctxd https://github.com/owner/repo/pull/123
ctxd https://github.com/owner/repo/pull/123 -o pr-123.md
ctxd -O https://github.com/owner/repo/pull/123

Options

Option Description
-d, --diff-mode full|compact|stat Diff output mode (default: compact)
--clean-body / --no-clean-body Strip bot-injected HTML noise from PR body (default: on)
--no-bots Drop bot-authored reviews and comments (default: keep all bots)

Slack Thread

Export a full Slack thread with username resolution and attachments.

Prerequisites

Requires a Slack User Token (xoxp-...) with the following scopes:

  • channels:history, groups:history, im:history, mpim:history
  • users:read
  • files:read (if downloading attachments)

Obtain at: api.slack.com/apps → Your App → OAuth & Permissions → User Token.

Configure the token (pick one):

# Option 1: Environment variable
export SLACK_TOKEN="xoxp-..."

# Option 2: Config file
mkdir -p ~/.config/ctxd
echo 'SLACK_TOKEN=xoxp-...' >> ~/.config/ctxd/config

Usage

# New URL format
ctxd https://app.slack.com/client/T.../C.../thread/C...-1234567890.123456

# Archive URL format
ctxd https://your-workspace.slack.com/archives/C.../p...?thread_ts=...

When you copy a link to a specific reply (archive URL with ?thread_ts= where the path ts differs from the thread root), ctxd fetches the entire thread but highlights the focused message — a **Focused Message:** line in the header and a ▶ marker on the corresponding message in the conversation flow.

Options

Option Description
--download-files Download attachments next to the output as attachments/IMG_{file_id}.{ext} (e.g. IMG_F0AAAAAAA1.png); requires -o/-O and the files:read scope on the Slack token
--raw Keep original Slack mrkdwn markup

Each attachment line keeps the Slack permalink (the address a human opens) and, when the file was downloaded, appends — saved: attachments/IMG_<id>.<ext> so the local copy is discoverable from the export. The run summary reports how many files were downloaded, or how many were left behind when the flag is not used.

Confluence

Export Confluence pages to Markdown. By default prints a single page to stdout; pass -r / -i with -o <dir> (or -O) to opt into recursive export with images.

Prerequisites

Requires an Atlassian API Token. Obtain at: id.atlassian.com/manage-profile/security/api-tokens.

Configure (all three values are required). Pick one:

# Option 1: Environment variables
export CONFLUENCE_BASE_URL="https://your-site.atlassian.net"
export CONFLUENCE_EMAIL="you@example.com"
export CONFLUENCE_API_TOKEN="your-token"

# Option 2: Config file (recommended for persistent use)
mkdir -p ~/.config/ctxd
cat >> ~/.config/ctxd/config <<'EOF'
CONFLUENCE_BASE_URL=https://your-site.atlassian.net
CONFLUENCE_EMAIL=you@example.com
CONFLUENCE_API_TOKEN=your-token
EOF
chmod 600 ~/.config/ctxd/config

Environment variables take precedence over the config file, so CI and one-off overrides work unchanged. If the file is readable by group/others, ctxd prints a one-shot stderr warning with the exact chmod 600 command to fix it.

Usage

# Default: single page to stdout
ctxd https://your-site.atlassian.net/wiki/spaces/SPACE/pages/123456

# Tiny link (short URL) — resolved via authenticated redirect
ctxd https://your-site.atlassian.net/wiki/x/ABC123

# Recursive export with images, to an explicit directory
ctxd https://your-site.atlassian.net/wiki/spaces/SPACE/pages/123456 -r -i -o ./output

# Or let ctxd pick the output directory name
ctxd https://your-site.atlassian.net/wiki/spaces/SPACE/pages/123456 -r -i -O

Note: -r / -i / --all-attachments require -o <dir> or -O (Confluence writes a directory tree / images to disk). Tiny links: /wiki/x/<token> URLs are followed once (after auth) and rewritten to the canonical long URL before any page fetch; -O auto-output names them confluence-<token> since the real page id isn't known at filename time.

Options

Option Description
-r, --recursive / --no-recursive Include child pages (default: off)
-i, --include-images / --no-include-images Download referenced images (default: off)
--all-attachments Download all attachments (default: only referenced images)
--debug Save raw HTML for debugging

Images that are not downloaded are not dropped: they link to the attachment's REST download URL, which works with the same API token (curl -u "$CONFLUENCE_EMAIL:$CONFLUENCE_API_TOKEN" -L <url>) and embeds no expiring token. The run summary reports how many images were left remote.

Jira

Export full Jira issue content (description, comments, custom fields, attachments).

Prerequisites

Shares authentication with Confluence (see above), so the same config applies. Jira also supports --debug for raw HTML saving.

Usage

ctxd https://your-site.atlassian.net/browse/PROJECT-123
ctxd https://your-site.atlassian.net/browse/PROJECT-123 -o issue.md

# Download every attachment next to the output file
ctxd https://your-site.atlassian.net/browse/PROJECT-123 --all-attachments -O

# Only the images the issue body actually references
ctxd https://your-site.atlassian.net/browse/PROJECT-123 -i -O

Rich-text custom fields are exported as Markdown. Serializable plain fields (strings, numbers, booleans, simple lists/dicts) are exported too. Unsupported nested objects are omitted with a stderr warning and a note in the run summary — never dropped silently.

Attachments

Every issue export ends with an Attachments section listing filename, MIME type and size, so nothing is invisible even when no file is downloaded. Attachments are only written to disk when -i or --all-attachments is passed:

Option Description
-i, --include-images Download images referenced from the issue body (default: off)
--all-attachments Download every attachment on the issue (default: off)

Downloaded files go to <output-stem>_attachments/, named <attachment-id>-<filename> so same-named uploads never overwrite each other. Attachment URLs in the description, comments and custom fields are rewritten to the local relative paths; anything that failed to download keeps its original remote URL.

Note: -i / --all-attachments require -o <file> or -O. When an issue has attachments and neither flag is set, the run summary says how many were skipped and which flag to use.

Completeness summary and manifest

Every run prints a one-line completeness summary to stderr (always visible, including under -q):

ctxd summary: source=jira | fetched=1 | rendered=1 | artifacts=1

Counts cover source resources fetched/rendered, artifacts written, plus skipped / failed / truncated when non-zero. Free-form notes (e.g. omitted custom fields, failed downloads) are listed under the line.

When writing to a file or directory, a machine-readable manifest is written next to the output:

Output mode Manifest path
Single file (-o issue.md / -O) issue.md.manifest.json
Directory (Confluence tree) <dir>/manifest.json

Manifest JSON mirrors the summary (counts, notes, per-item status). Cross-source recursion merges child counts into the root summary; child content is embedded in the same artifact, so artifacts stays 1.

Output integrity and limits

  • Atomic writes — text and binary files are written via temp file + rename to avoid truncated artifacts on interrupt.
  • --max-chars — truncates at a newline boundary, closes open code fences, and appends a truncation notice. Hard cap: output length never exceeds the limit. Default applies to stdout; pass the flag explicitly to also cap file output.
  • Attachment size caps — streamed downloads enforce per-file (--max-file-size) and per-run (--max-run-size) budgets shared across the recursion tree. Exceeded downloads warn and count as failed rather than filling the disk.

Cross-source recursion

By default, cross-source recursion is off (--recurse-depth 0). Opt in with --recurse-depth 1 or 2 to have ctxd scan the rendered output for supported URLs (Slack, GitHub PR, Confluence, Jira) and fetch them too, appending the results as a labelled appendix. This means a Slack thread that links a Jira issue and a GitHub PR pulls all three in one command — no follow-up fetches needed.

# Recursion off (default) — primary URL only
ctxd https://app.slack.com/client/.../thread/...

# Enable recursion (depth=1, auto-expands linked supported URLs)
ctxd <url> --recurse-depth 1

# Deeper recursion (max 2)
ctxd <url> --recurse-depth 2

Key behaviours:

  • Deduplication: the same URL is never fetched twice within one run.
  • Cap: at most 5 child URLs are expanded per level (prevents Jira issue-link explosions).
  • Graceful skip: if a child URL lacks credentials (e.g. a Slack thread links a Confluence page but no Confluence token is configured), the appendix notes the skip instead of failing.
  • Confluence directory export: recursion is disabled when using -o/-O with Confluence (which writes a page tree to disk); use stdout for recursion.

Performance

Scenario Baseline Post-opt Improvement
Slack thread 9.09s 1.61s 82.3%
Confluence single page + image 1.88s 1.74s 7.4%
Confluence recursive + images 27.13s 4.04s 85.1%
GitHub PR 6.75s 4.15s 38.5%

License

MIT

About

Turn work URLs into LLM-ready Markdown. CLI to export GitHub PRs, Slack threads, Confluence pages, and Jira issues. Claude Code & Codex skill included.

Topics

agent-toolsclaude-codeclicodexgithub-prllm-toolsmarkdown-exportskillsslack

Resources

Readme
MIT license
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Used by

Contributors

Languages