Actions Reference

All composite actions are referenced as fetch-rewards/buck-bronson/.github/actions/<action>@v1 (pinned to the floating major version tag).

`bucky-execute`

Runs Claude against a GitHub issue with AWS Bedrock auth. Supports a flexible mode system: empty/default mode lets Claude decide what to do, plan mode restricts to planning only. Prompt templates live in prompts/ with {{PLACEHOLDER}} substitution.

Input	Required	Description
`mode`	No	Empty (Claude decides) or `plan` (planning only)
`issue-number`	Yes	GitHub issue number
`claude-args`	No	Additional arguments to pass to Claude CLI
`additional-prompt`	No	Extra instructions appended to the base prompt
`additional_permissions`	No	Extra GitHub permissions beyond defaults
`plugins`	No	Comma-separated plugins from buck-bronson marketplace
`sourcegraph-token`	No	Sourcegraph API token
`rollbar-token`	No	Rollbar access token
`posthog-token`	No	PostHog API token (web presets only)
`github-token`	Yes	GitHub App token with cross-repo access
`oidc-role`	Yes	AWS IAM role ARN for OIDC authentication

Prompt placeholders: {{REPOSITORY}}, {{ISSUE_NUMBER}}, {{ADDITIONAL_PROMPT}}

Multi-repo prompts (in .github/actions/bucky-execute/prompts/) use additional placeholders: {{ISSUE_REPO}}, {{REPO_LIST}}, {{REPO_PATHS}}

`determine-mode`

Detects whether Bucky should run and extracts mode from comment content and event type.

Input	Required	Description
`mode`	No	Explicit mode override — skips event detection

Output	Description
`mode`	Empty (Claude decides) or `plan` (planning only)
`should-run`	`true` if Bucky should run
`is-issue`	`true` if the event is on an issue (not a PR)

Detection logic:

@bucky /plan → plan mode
@bucky (without /plan) → empty mode (Claude decides)
Explicit mode input → uses that value with should-run=true

`bucky-code-review`

AI code review on a pull request.

Input	Required	Description
`claude-args`	No	Additional arguments to Claude CLI
`plugins`	No	Comma-separated plugins
`sourcegraph-token`	No	Sourcegraph API token
`rollbar-token`	No	Rollbar access token
`posthog-token`	No	PostHog API token
`github-token`	Yes	GitHub App token
`oidc-role`	Yes	AWS IAM role ARN

`tag-bucky`

Handles @bucky ad-hoc mentions on PRs.

Input	Required	Description
`claude-args`	No	Additional arguments to Claude CLI
`prompt`	No	Custom prompt (default: use comment content)
`additional_permissions`	No	Extra GitHub permissions (default: `actions: read`)
`plugins`	No	Comma-separated plugins
`sourcegraph-token`	No	Sourcegraph API token
`rollbar-token`	No	Rollbar access token
`posthog-token`	No	PostHog API token
`github-token`	Yes	GitHub App token
`oidc-role`	Yes	AWS IAM role ARN

`resolve-preset`

Reads presets.json and resolves runtime configuration for a preset.

Input	Required	Description
`preset`	Yes	Preset name (`web`, `golang`, `none`)
`github-token`	Yes	Token with access to buck-bronson

Output	Description
`plugins`	Plugin list (multiline, one per line)
`setup-action`	Setup action name (`setup-web`, `setup-golang`, or empty)
`has-posthog-token`	Whether this preset requires the PostHog token

`setup-web`

Sets up Node.js development environment.

Input	Required	Description
`npm-registry`	Yes	NPM registry URL
`buf-access-token-secret-arn`	Yes	ARN for Buf access token secret
`web-auth-cookies`	No	Base64-encoded cookie JSON for browser auth

What it does: Detects pnpm vs npm, detects .node-version vs .nvmrc, installs dependencies, then calls setup-playwright.

`setup-golang`

Sets up Go development environment.

Input	Required	Description
`go-proxy`	Yes	Artifactory proxy URL

What it does: Sets GOPROXY env var, uses mise for version management, runs go mod download.

`setup-playwright`

Installs browser automation tooling.

Input	Required	Description
`cookies`	No	Base64-encoded JSON array of cookies

What it does:

Installs Playwright Chromium (npx playwright install --with-deps chromium)
Installs playwright-cli globally (npm install -g @playwright/cli)
Writes .playwright/cli.config.json with headless Chromium and ignoreHTTPSErrors
If cookies provided, writes a Playwright storage state file to /tmp/playwright-storage-state.json

`report-session-outcome`

Collects Claude session artifacts and POSTs results back to the agent callback endpoint.

Input	Required	Description
`callback-url`	Yes	Agent callback endpoint URL
`session-id`	Yes	Workflow session UUID
`issue-number`	Yes	GitHub issue number
`mode`	No	Session mode (empty or `plan`)
`status`	Yes	Job status (pass `job.status`)

Artifacts collected: Recent bot comment, linked PR, labels (needs-review, needs-attention).

Authentication: Gets an OIDC token with the callback URL as audience. Retries 3 times with exponential backoff.