solar-code

Reusable Solar protocol for code modifications in planet-operated repos. Use when an intention (RFC, task, direct instruction) must be converted into a local, human-reviewable code change. Covers tria

SKILL.md

GitHub

Solar Code

Purpose

Convert an intention into a local executable code change without reinventing the rules each time. Provides a canonical flow and repo adoption contract for repos operated by planets.

When to Use

Use this skill when:

  • A task requires modifying code in a planet-operated repo target.
  • A change needs a defined triage level before touching files.
  • A planet needs to declare its repo policy (allowlist, restrictions, checks).

Do not use for changes inside core/skills/ — those are governed by solar-skill-creator.

Do not use for:

  • PR creation, git push, or GitHub review workflows.
  • CI/CD pipeline changes.
  • Async workers or queues.
  • Multi-repo architectural decisions (use an RFC first).

Required MCP

None

Triage

Classify the change before acting:

Level Description Required artifact
Micro change 1-2 files, low risk Clear instruction + repo checks
Standard change Feature or fix with relevant context Lightweight task spec (Markdown)
Multi-repo / high risk Touches multiple repos or has strategic impact RFC + task spec + prior review

Workflow

  1. Triage — classify the change level (micro / standard / multi-repo).
  2. Load repo policy — read the target repo's CONTRIBUTING.md file before writing anything.
  3. Write task spec (if standard or above) — use references/task-spec.md as the single source of truth for structure and allowed sections.
  4. Apply change locally — edit files in the working tree; do not push.
  5. Update completion evidence — after implementation, record only the facts needed for review and traceability.
  6. Run checks — only commands declared in the repo policy allowlist.
  7. Human review in IDE — surface the diff; human decides to keep or discard.

Default mode: local-review. Branch, push, and PR are optional layers added only when there is evidence they are needed.

Authority (supervised autonomy)

  • Clear local solar-code work under an explicit user mandate is A2 implicit (scoped working-tree edit + IDE review). Do not ask a second “approve?” for that scoped edit.
  • Push, PR, tag, and release are never A2 implicit — require A2 formal (and treat irreversible release paths as A4 escalation when appropriate).
  • If triage level grows beyond the original mandate, re-triage and get explicit approval before continuing (new A2).

Failure protocol

  • If a required check fails: stop, report the failure, do not proceed.
  • If the repo policy is missing: ask the user to declare one before writing files.
  • If the change scope grows beyond the original triage level: re-triage and get explicit approval before continuing.

Repo adoption contract

Each repo target must declare a policy file at its root. Format: CONTRIBUTING.md. This file defines the "rules of engagement" for both humans and AI agents.

For code repositories adopting this skill, the recommended split is:

  • AGENTS.md — planet governance, routing, architecture ownership.
  • CONTRIBUTING.md — repo policy, checks, restrictions, task-spec location.
  • docs/tasks/ — standard-or-higher task specs unless the repo declares a different path.

References

  • references/task-spec.md — canonical task spec template with optional completion evidence.
  • references/repo-policy.md — repo policy template (rename to CONTRIBUTING.md when adopting a repo).
  • references/local-review-guide.md — how to use local-review mode.