goldband

Files (200)

• .env.example
• .gitattributes
• .gitignore
• AGENTS.md
• ARCHITECTURE.md
• CLAUDE.md
• ETHOS.md
• LICENSE
• README.md
• SKILL.md
• SKILL.md.tmpl
• VERSION
• openai.yaml · agents/
• dev-setup · bin/
• dev-teardown · bin/
• goldband · bin/
• goldband-analytics · bin/
• goldband-artifacts-init · bin/
• goldband-artifacts-url · bin/
• goldband-brain-enqueue · bin/
• goldband-brain-restore · bin/
• goldband-brain-sync · bin/
• goldband-brain-uninstall · bin/
• goldband-builder-profile · bin/
• goldband-codex-probe · bin/
• goldband-config · bin/
• goldband-cross-review · bin/
• goldband-developer-profile · bin/
• goldband-diff-scope · bin/
• goldband-env · bin/
• goldband-global-discover.ts · bin/
• goldband-ios-qa-daemon · bin/
• goldband-ios-qa-mint · bin/
• goldband-jsonl-merge · bin/
• goldband-knowledge · bin/
• goldband-knowledge-review · bin/
• goldband-knowledge.ts · bin/
• goldband-learnings-log · bin/
• goldband-learnings-search · bin/
• goldband-model-benchmark · bin/
• goldband-next-version · bin/
• goldband-open-url · bin/
• goldband-paths · bin/
• goldband-platform-detect · bin/
• goldband-pr-title-rewrite.sh · bin/
• goldband-question-log · bin/
• goldband-question-preference · bin/
• goldband-repo-mode · bin/
• goldband-review-log · bin/
• goldband-review-read · bin/
• goldband-session-update · bin/
• goldband-settings-hook · bin/
• goldband-slug · bin/
• goldband-team-init · bin/
• goldband-telemetry-log · bin/
• goldband-timeline-log · bin/
• goldband-timeline-read · bin/
• goldband-uninstall · bin/
• goldband-update-check · bin/
• goldband-upgrade-context · bin/
• …and 140 more

SKILL.md (first lines)

<!-- AUTO-GENERATED from SKILL.md.tmpl and goldband.manifest.json. Do not edit. -->

# Goldband capability router

Use the exact interface `$goldband <capability> <action>`. Old workflow names are
not aliases and must fail as unknown input.

For an empty invocation, show this generated capability menu and stop:

<!-- AUTO-GENERATED from goldband.manifest.json. Do not edit. -->
- `$goldband review code` — Independent review with a selected engineering lens.
- `$goldband investigate code` — Find and verify the root cause of a failure.
- `$goldband qa app` — Verify product behavior with explicit evidence.
- `$goldband plan create` — Create, expand, or tune an implementation plan.
- `$goldband design consult` — Define, explore, and prototype product design.
- `$goldband safety guard` — Apply or remove workflow safety boundaries.
- `$goldband context restore` — Save, restore, and summarize working context.
- `$goldband knowledge recall` — Recall Goldband knowledge.
- `$goldband benchmark workflow` — Measure workflow or model performance.
- `$goldband document generate` — Create product documentation and publication artifacts.
- `$goldband system health` — Inspect or maintain the Goldband installation.
- `$goldband ios qa` — Maintain and validate iOS-specific product behavior.

## Route and load

1. Select one capability/action from the menu using the user's requested outcome.
   Do not force a workflow when ordinary conversation is sufficient.
2. Resolve the active Goldband runtime root from `GOLDBAND_ROOT`, or from the
   directory containing this `SKILL.md`.
3. Read `workflows/<capability>/<action>.workflow.md` completely. That file is
   the task contract. Do not search for nested workflow `SKILL.md` entrypoints.
4. Keep the working prompt limited to the requested goal, relevant context,
   hard boundaries, and verification. Let the model choose its reasoning and
   tools unless the contract requires a deterministic process.
5. Require approval before an outward-facing or irreversible action. Verify the
   requested outcome before reporting completion.

## Human decisions

- Ask only when user authority is required, or two plausible answers would change the target, behavior, or external effect and current evidence or preferences cannot resolve intent.
- Batch related decisions when they can be answered together; split only when an earlier answer changes the next question, risk level, or required evidence.
- Tool schemas and UI own question shape, option labels, validation, and persistence. Prompts should provide only concise decision context.
- Avoid prompt-owned formats: prompt-owned formatting rubrics, scores, word-count rules, per-finding question rules.

Load on-demand manuals only for their generated selectors:

<!-- AUTO-GENERATED from goldband.manifest.json. Do not edit. -->
- Read `manuals/browser.md` only for `qa/app`.

Load host-specific setup only when the selected contract explicitly points to
it.

If the capability/action is unknown or its contract is missing, stop with an
explicit error. Do not redirect to a historical name or fabricate a workflow.

From leo110047/goldband, MIT.

More from leo110047/goldband