Agent plugins

Repowise ships two agent plugins. Both bundle the MCP server, lifecycle hooks, and a set of skills into a single install, so you do not wire up each piece by hand.

Both are AGPL-3.0 and live in the OSS repo under plugins/.

Claude Code plugin#

/plugin marketplace add repowise-dev/repowise
/plugin install repowise@repowise

That is the whole install. It brings the MCP server, 18 slash commands, six skills, and three hooks.

Slash commands#

Explicit, you type them. Each maps to a CLI command and formats the result for reading rather than parsing.

Skills#

Implicit, they activate on their own when the work matches. Each checks for a .repowise/ directory first and stays out of the way if the repo is not indexed, so installing the plugin globally does no harm in unindexed repos.

Hooks#

Three, all calling the repowise-augment entry point:

Claude Code also gets search-result enrichment on Grep and Glob, which the Codex setup does not attempt. Full detail in Agent hooks.

The claude_cli provider#

Separate from the plugin, and pointing the other way. The plugin lets Claude Code call repowise. This lets repowise call Claude Code: page generation runs through your local Claude Code CLI, so a Pro, Max, Team or Enterprise subscription does the work instead of a metered ANTHROPIC_API_KEY.

repowise init --provider claude_cli --yes

For an index that already exists:

REPOWISE_PROVIDER=claude_cli repowise generate --unwritten

To persist the choice, put it in .repowise/config.yaml:

provider: claude_cli
model: claude_cli/claude-haiku-4-5

It needs the Claude Code CLI installed and authenticated once with claude login. Any plan that can run claude -p works. Repowise never sees your credentials: the CLI holds them, exactly as with codex_cli and opencode.

claude_cli/claude-haiku-4-5 is the default model. The claude_cli/ prefix is optional on input, so --model claude-opus-4-6 and --model claude_cli/claude-sonnet-4-6 both work and both round-trip through the config file.

Caveats worth reading before you commit to it:

• It is slower than an API key. Each page is a full CLI process. Budget roughly 9k output tokens and a few minutes per page, drawn from the same limits as your interactive Claude Code sessions. On a large repo, scope a run with --path rather than regenerating everything.
• Concurrency is bounded at 4. Subscription limits are per account, and too much concurrency trips the limit and fails the run. The provider runs at most 4 CLI processes at a time, whatever --concurrency says; a lower --concurrency still lowers it. Raise or lower the cap with REPOWISE_CLAUDE_CLI_CONCURRENCY=2 repowise generate --unwritten. Going above 4 is reasonable on a larger allowance, and it is also the fastest way to trip the account limit mid-run.
• Reasoning modes do not map one for one. low, medium, high, xhigh and max pass through to the CLI's --effort; auto omits the flag. Repowise's off, none and minimal have no equivalent, so they are warned about and ignored rather than failing the run.
• Cost is recorded as $0.00, because billing sits with the subscription. The CLI's own total_cost_usd is still kept under usage.reported_cost_usd if you want to audit it.

The subprocess runs in a fresh empty scratch directory, not in your repository, so Claude Code does not auto-discover your CLAUDE.md and inject project instructions into every page prompt. Tools are disabled, turns are capped at one, and --strict-mcp-config stops the subprocess loading MCP servers from your config.

An agent inside a Claude Code session can run repowise init --provider claude_cli; the nested claude -p calls work, and auth and backend settings such as CLAUDE_CODE_USE_BEDROCK pass through.

Detected, never picked for you. repowise init reads claude auth status (and codex login status, opencode auth list) to see how each CLI is billed, and the plan screen shows that in place of a price, for example "uses your plan quota, about 14 model calls, 4 at a time, about 4 min". A --yes run never switches to one: it stays free and prints the command that would use it. A CLI signed in without a paid plan is listed but never recommended. Codex signed in through ChatGPT is listed but not offered, since its status cannot show whether the plan is active; Codex with an API key is offered, billed per token.

Codex plugin#

Repowise supports Codex three separate ways, and they are worth keeping apart:

1. Project setup for Codex MCP and lifecycle hooks, via repowise init --codex.
2. The codex_cli provider, which generates wiki pages through your authenticated Codex CLI subscription instead of an API key.
3. The plugin, which bundles MCP, hooks, and skills.

Prerequisites#

npm install -g @openai/codex
codex login
codex login status

Repowise checks codex --version and codex login status. When both succeed, interactive repowise init offers Codex project setup. Non-interactive runs need --codex explicitly; --no-codex skips the prompt.

Project setup#

repowise init --codex

This writes project-local files, not your global ~/.codex/config.toml. It merges into .codex/config.toml:

[mcp_servers.repowise]
command = "repowise"
args = ["mcp"]
cwd = "/absolute/path/to/repo"
startup_timeout_sec = 20

[features]
hooks = true

The server runs repowise mcp with no path argument. In no-path mode it walks upward from the current directory to the nearest initialized .repowise repo. Smoke check with codex mcp list.

Installing the plugin#

The Codex plugin installs from a local marketplace rather than a public one, so you need a clone of the repo:

git clone https://github.com/repowise-dev/repowise
cd repowise
codex plugin marketplace add .
codex
/plugins

It bundles the MCP server, lifecycle hooks, and the same six skills in Codex-neutral form. It adds no slash commands, because Codex has no equivalent surface.

Codex hooks#

Written to .codex/hooks.json, not inline [hooks] tables:

The codex_cli provider#

Worth knowing about even if you skip the plugin. It routes page generation through your Codex subscription instead of a metered API key:

repowise init --provider codex_cli --codex --yes

Repowise treats codex_cli/* cost as $0.00, because subscription billing happens outside its API pricing. --reasoning minimal maps to Codex model_reasoning_effort="low"; low, medium, high and xhigh pass through when the selected model advertises them. off and none are not supported by this provider.

AGENTS.md#

repowise init --codex generates a managed AGENTS.md by default, the Codex counterpart to the generated CLAUDE.md. repowise update refreshes it when editor_files.agents_md is enabled or --agents is passed. The repowise section sits between managed markers and your own content outside those markers is preserved.

repowise init --no-agents      # skip it
repowise init --agents         # force it
repowise update --agents       # refresh on update

The copilot_cli provider#

Page generation can also run through a logged-in GitHub Copilot CLI, so a Copilot subscription does the work with no API key:

repowise init --provider copilot_cli --yes
REPOWISE_PROVIDER=copilot_cli repowise generate --unwritten   # an existing index

It needs the copilot CLI on PATH, signed in once. Repowise runs it with tools, built-in MCP servers and custom instructions turned off, and records the cost as $0.00 because billing sits with the subscription. REPOWISE_COPILOT_CLI_CONCURRENCY sets how many processes run at once. Like the other agent CLIs it has no embedder, so semantic search needs an embedder key of its own.

Which to install#