Recommended third-party companions¶

Optional tools that pair well with ADHD Progress Hub. Hub does not require them. They are independent projects: Hub does not ship, warranty, or auto-update them.

Formerly titled “coding companions.” The published /coding-companions/ URL is kept stable; the set now includes writing and workflow tools as well as coding helpers.

For the Cursor Marketplace client plugin (skills + rule + BYO MCP), see adhd-hub-cursorskill. Hub remains the skill source of truth; sync opens a plugin PR — Cursor plugin skill sync.

Advanced companions¶

Choose only what solves a real problem¶

Installing every companion is not recommended. Start with the Hub by itself, then add a tool only when it fixes a specific workflow problem.

Examples:

• need more action-first replies → i-have-adhd;
• want structured agent workflows → Superpowers;
• want less unnecessary abstraction → Ponytail;
• need codebase/library context → Graphify or Context7;
• need browser verification → agent-browser;
• want quieter shell output → RTK;
• want natural prose cleanup → Humanizer;
• need heavier semantic code navigation → Serena.

After adhd-hub connect or adhd-hub doctor, the CLI prints a short companions checklist. Enable Hub-installable ones in Hub Settings → Coding agents → Helpful extras, or pass install flags:

adhd-hub connect /path/to/project \
  --agents cursor,codex,claude \
  --with-i-have-adhd \
  --with-superpowers \
  --with-ponytail \
  --with-graphify \
  --with-rtk \
  --with-context7 \
  --with-agent-browser \
  --with-serena \
  --with-humanizer

Use the same --agents list as Hub MCP/skills. With no agents selected, the CLI links this guide instead of guessing per-agent setup.

You can also put flags in ADHD_HUB_CONNECT_FLAGS for the install one-liner (including --with-ponytail and --with-humanizer).

What Hub automation can do¶

Hub can opt-in install i-have-adhd, Superpowers, Ponytail, Graphify, RTK, Context7, agent-browser, Serena, and Humanizer through Settings or --with-* flags.

Important differences:

• Superpowers and Claude Code Ponytail are best-effort because some hosts require an in-app plugin install.
• Context7 and Serena merge MCP configuration.
• agent-browser installs its CLI and skill.
• Humanizer installs the skills.sh skill only.
• Optional companion failures are warnings; they never fail Hub connect.

Hub passes --yes to npx as well as -y to the skills CLI, so npm's first-use package prompt does not interrupt these installs. With --agents '*', some agents do not support global skills; partial installations are reported as warnings even if the upstream CLI exits zero. Install those agents' skills at project scope or select only the agents you use. Manual plugin and hook-trust steps remain under Needs attention until completed in the relevant agent. Each skills installation has a five-minute timeout; stalled optional installs warn so Hub connect can continue. Retry the command after resolving the stall.

i-have-adhd¶

Category: ADHD & interaction License: MIT What it does: Shapes assistant replies for action-first, numbered steps (opt-in skill; no ADHD diagnosis required). Credits The Adult ADHD Tool Kit (Ramsay & Rostain). Privacy: Markdown skill / plugin rules — no product telemetry.

Install (skills CLI)¶

# One or more skills.sh agent ids (Hub maps claude → claude-code):
npx skills add ayghri/i-have-adhd -g -y -a cursor -a codex -a claude-code

# Or every skills.sh agent:
npx skills add ayghri/i-have-adhd -g -y --agent '*'

Then invoke /i-have-adhd (or $i-have-adhd in Codex) in a new session. Full platform matrix (Claude plugin, Gemini, Pi, …): upstream INSTALL.md.

Superpowers¶

Category: Agent workflow License: MIT What it does: Agent workflow skills (brainstorm → plan → TDD → review). Privacy: Local skills/plugin; see upstream for any optional telemetry (visual companion). Install (opt-in): --with-superpowers / Settings checkbox. Hub prints harness-specific steps and best-effort CLI install where available (e.g. Gemini extensions). Most harnesses still need an in-app plugin install — see obra/superpowers.

Ponytail¶

Category: Code quality License: MIT Upstream: DietrichGebert/ponytail What it does: Encourages agents to prefer the smallest correct implementation, avoid premature abstractions and review changes for unnecessary complexity. Privacy / execution: Local plugin, hooks, or rules. Upstream may inject a ruleset via agent lifecycle hooks. Hub does not send Ponytail traffic anywhere. Review upstream before enabling. Independence: Completely optional and independently maintained. Hub does not vendor Ponytail source and never makes it mandatory.

Install vs activation¶

• Install = place the plugin/hooks/extension for your agent (what --with-ponytail does).
• Activation / mode = upstream-controlled (lite / full / ultra / off). Hub does not force a strictness level. Upstream’s default after install is typically full unless you set PONYTAIL_DEFAULT_MODE or ~/.config/ponytail/config.json. Switch with /ponytail lite|full|ultra|off (or plain messages in Cursor hooks).

Hub opt-in install¶

--with-ponytail / Settings checkbox. Hub follows upstream recipes per selected agent (not skills.sh as the primary path — Ponytail’s README documents plugins/hooks/extensions):

Manual upstream install¶

See the upstream README Install section for Claude, Codex, Cursor, Gemini, and other hosts. Failures warn and never fail Hub connect.

Graphify¶

Category: Code intelligence License: Apache-2.0 (older portions also under MIT; see upstream NOTICE) What it does: Builds a local knowledge graph of your repo so agents can query / path / explain instead of blind grepping. Privacy: Code AST extraction is local. No product telemetry. Docs/media semantic passes use your assistant or a configured API key. Commercial graphify.com is separate from the open-source CLI.

Install¶

uv tool install graphifyy   # package name is graphifyy; CLI is still `graphify`

Register for the agents you actually use (examples):

Then in a project: /graphify . (PowerShell: graphify .). Upstream: Graphify README.

Context7¶

Category: Code intelligence License: MIT What it does: Up-to-date library docs via MCP/CLI. Privacy: Queries go to Context7’s service; may need an API key — see upstream. Install (opt-in): --with-context7 / Settings checkbox. Hub merges an MCP stdio entry (npx -y @upstash/context7-mcp) into selected agent configs. Optional API key improves rate limits — see upstash/context7.

agent-browser¶

Category: Execution & verification License: Apache-2.0 (confirmed from upstream LICENSE) What it does: Browser automation for agents to verify UI. Privacy: Drives a local browser; review upstream before enabling. Install (opt-in): --with-agent-browser / Settings checkbox. Hub runs npm install -g --allow-scripts=agent-browser agent-browser, agent-browser install, and the skills installer for selected agents. The npm option permits this package's postinstall script without changing your global npm configuration. Hub then opens a local about:blank page in a temporary session with an empty browser configuration and no AGENT_BROWSER_* overrides, and attempts to close that session. A download alone does not count as a verified launch. Launch and cleanup failures warn; cleanup warnings include the session's close command for recovery. Hub does not install system packages, change the browser sandbox, or attach this check to an existing browser. See vercel-labs/agent-browser.

RTK¶

Category: Execution & verification License: Apache-2.0 What it does: Compresses shell/tool output before your agent reads it; optional hooks rewrite bash commands. Privacy: Telemetry is opt-in (see upstream TELEMETRY.md / README). Leave it disabled unless you consent at RTK’s prompt. Hooks change how agents run shell commands — review before enabling.

Install binary¶

adhd-hub connect . --with-rtk --agents cursor,codex (or the Settings checkbox with agents selected) tries Homebrew when brew is on PATH, otherwise runs the upstream installer below (needs curl). The binary install still runs when --with-rtk is set without --agents; rtk init needs agents. The script installs into ~/.local/bin. RTK hooks call bare rtk, so that directory must be on your PATH; Hub warns with the exact export PATH=… line when it is not.

# macOS (Homebrew) — preferred when brew is available
brew install rtk

# Linux / macOS script (review before piping) — used by Hub when brew is absent
curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh

Windows: download a release zip from rtk releases, put rtk.exe on PATH (Hub does not auto-download on Windows).

Init for your agents¶

Restart the coding agent after init. Upstream: RTK README.

Humanizer¶

Category: Writing License: MIT Upstream: blader/humanizer What it does: Rewrites AI-generated prose to sound more natural while preserving the original meaning. Best suited to documentation and user-facing writing. Built on Wikipedia’s Signs of AI writing. Privacy: Local skill; runs in your agent when you invoke it. No Hub post-processing layer. Independence: Completely optional and independently maintained. Hub does not vendor Humanizer source.

Behaviour (important)¶

Hub only installs the skill. It does not:

• auto-rewrite every agent response
• run Humanizer against source code, YAML/JSON/config, generated structured data, shell commands, or internal agent state

Invoke on demand (/humanizer, or ask the agent to humanize a prose draft). When pointed at a file, upstream edits prose and leaves code/data/frontmatter alone — still use it intentionally on writing, not as a blanket pass.

Install (skills CLI / Hub opt-in)¶

# Hub:
adhd-hub connect . --agents cursor,codex,claude --with-humanizer

# Manual (skills.sh):
npx skills add blader/humanizer -g -y -a cursor -a codex -a claude-code
# or: npx skills add blader/humanizer -g -y --agent '*'

Claude Code can also use the upstream plugin marketplace (/plugin marketplace add blader/humanizer then /plugin install humanizer@humanizer) — see blader/humanizer.

Serena (advanced)¶

Category: Code intelligence License: MIT What it does: Semantic code navigation / editing via MCP. Heavier setup than the companions above—use when you want LSP-style repo intelligence. Privacy: Typically local LSP/MCP; confirm upstream if using remote options. Install (opt-in): --with-serena / Settings checkbox. Hub runs uv tool install -p 3.13 serena-agent, best-effort serena init, and merges MCP launch config for selected agents. See oraios/serena.

Ethics and expectations¶

• These are recommendations, not Hub dependencies.
• Prefer linking and opt-in CLI flags over silent installs.
• Hub install paths: --with-i-have-adhd / --with-superpowers / --with-ponytail / --with-graphify / --with-rtk / --with-context7 / --with-agent-browser / --with-serena / --with-humanizer (and matching Settings checkboxes). All are opt-in; failures warn and never fail Hub connect.
• Respect each project’s license and attribution when redistributing (Hub does not vendor their code).
• For RTK, prefer the README/TELEMETRY docs over any conflicting one-line disclaimer about defaults.