itismixian/orrery/skill-entry-router-v0-1

Project Orrery

Keep one traceable fact chain without making every task carry every Orrery operation.

Route before loading detail

• Existing-project maintenance or audit: obey the target AGENTS.md; read its mandatory entrances,

relevant State Docs, effective ADRs only when reasons are needed, and the active Implementation Plan before implementation. Do not load install, release, migration, or viewer references unless the task actually crosses those boundaries.

• New scaffold: read scaffold.md.
• Skill update, viewer upgrade, or compatibility work: read

release-channel.md and migration-contract.md.

• Architecture explanation or document-schema migration: read

architecture.md and migration-contract.md.

• Run or configure the observatory: read viewer.md.

Always preserve these boundaries

• Seed constrains durable principles; ADR records accepted choices and reasons; approved Design specifies

them; Plan records intent; code/configuration/assets/external state are implementation truth; State Docs report current facts; Validation records reproducible evidence.

• accepted is not implemented; a worktree change is not a commit; a commit is not a release.
• Never bulk-overwrite authored AGENTS.md or docs/. Default installation creates missing files only;

managed-tool upgrades require a reviewed dry run and backup.

• Never package credentials, ai-config.json, keyring contents, caches, .port, generated docs/_site/,

virtual environments, or user-specific benchmark output.

• Dashboards, AI answers, summaries and radar are projections, not new project facts.

Close maintenance work

Validate in proportion to the changed boundary and follow the target Plan/AGENTS commands. After real implementation or validation, update affected State Docs and PROGRESS, append DEVLOG, and refresh HANDOFF when the stopping point or risks changed. Report implementation, commit, push and release status separately.