Index of every page on this site: llms.txt
All skills
scode avatar

/scode-galaxy-brain

@03a9f7f
Run it yourself

Accomplish a goal by delegating suitable parts of the work to cost-effective models while the current session stays in charge of planning, quality gating, and all commit/PR management. Use when the user explicitly invokes scode-galaxy-brain, e.g. "Use scode-galaxy-brain for a goal", optionally with a prefer-gpt, prefer-claude, prefer-muse, or prefer-glm keyword and/or a request to work with concurrency. Also use when the user says "galaxy brain feedback: ..." to record feedback about how this skill performed. Once invoked, the skill stays active for the rest of the session — including across context compaction and resume — until the user expressly stops it, unless the invocation itself limited the scope up front.

  • 12 files
  • 82.3 KB
  • Updated last month
  • GitHub
Use this Skill: https://skilld.dev/gh/scode/dotfiles/scode-galaxy-brain

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SPEC.md

≈2k tokens on demand. Your agent reads this file only when SKILL.md points to it.

scode-galaxy-brain specification

Dependencies: scode-model-routing, scode-agent-delegation

NOTE: This file is binding on the skill's text. It is deliberately sparse: it records only requirements that have been stated as such, not a description of everything the skill does. Absence of an entry means the behavior is not yet specified, not that it is unspecified on purpose. When the skill and this file disagree, that is a bug in one of them; fix the skill or change this file in the same change, never leave them apart.

Requirements

  • The session in which the user invokes this skill is the orchestrator, regardless of harness, provider, model, or model tier. Routing a unit to a stronger model does not transfer that role, and missing delegation capabilities do not disqualify the session from it.
  • The skill's dependency set is the Dependencies: line above, and the dependency graph between skills is one-way: this skill may refer to a skill it depends on, and a skill it depends on never refers back to it. A dependency's text that points a reader here is a bug in the dependency; this skill's text that reaches into a dependency by any path other than the one below is a bug here. The shell-out skill is not a dependency of this skill: it is loaded by the delegation skill, and this skill's text names it only to say so.
  • Dependencies use the current harness's authorized skill loader or resource resolver, or, when no dedicated loader exists, the exact SKILL.md location its catalog or instructions supplies. On Codex, when no location is supplied, use ${CODEX_HOME:-$HOME/.codex}/skills/<name>/SKILL.md (the root verified for Codex 0.152). Known interfaces are examples, not an allowlist of harnesses. Missing filesystem metadata or an unfamiliar harness alone must not block loading or trigger a permission question; actual tool permissions remain binding. Never guess installation paths or URI schemes, assume a sibling dependency directory, or search other skill roots.
  • Identify the exact requested skill from its returned frontmatter, or the loader's reported identity when frontmatter is not exposed. Missing or conflicting identity fails loading. Read the skill in full and all sidecars the current step needs, using the resolver for that same skill, its reported base, or the directory of its supplied SKILL.md path. A filesystem base is unnecessary when a resolver addresses required resources or no sidecars are needed.
  • Truncated or elided output is incomplete delivery, not a terminal failure by itself. Recover omitted content through the tool's continuation, range reads, or full-output artifact tied to the same resource or result before acting. Unknown skills, denied access, invalid identity, unresolved required resources, or content that cannot be fully retrieved stop the affected operation and name the skill and failing path, URI, or tool. No permission bypass, remembered instructions, alternate copy, or similar skill may replace a failed load.
  • The loading text is the marked stanza in SKILL.md, one per dependency, checked against the canonical template in tests/skill_deps.rs. On omp, read resolves skill://<name> and skill://<name>/<relative-path>; these satisfy the same capability-based contract as other authorized loaders.
  • A same-named project-local skill may shadow the installed dependency when the harness selects it. This is accepted: the name check proves identity, not revision. Codex's fixed-path fallback does not discover project-local overrides.
  • Every choice of model, effort, and launch mechanism is a request to scode-model-routing naming a work profile with the request inputs that skill's SKILL.md lists, acted on through the answer shape it defines. This skill's text never names a model as a routing default, never carries an inventory or profile table, and never reads the model routing config file itself.
  • Every delegation goes through scode-agent-delegation, which generates the run id, owns the run directory, launches, runs the checkpoint protocol, and returns one verdict from its vocabulary. This skill's text carries no copy of the task-spec checklist, the checkpoint protocol, the gate's judging steps, or the addendum, and it acts on every verdict in that vocabulary through its action table; a verdict with no row is a bug here.
  • Multiple concurrent uses of scode-galaxy-brain by different orchestrators must not conflict through any state the skill itself maintains. Two or more orchestrating sessions may be running the skill at the same time — on the same machine, against the same repository, possibly in the same working tree — and every on-disk artifact the skill introduces for its own purposes (checkpoint files, scratch copies, prompt files, result and log files, trees it creates for isolated delegates) must be private to the orchestrator that created it. No orchestrator may remove, overwrite, or reinterpret another orchestrator's artifacts, and the skill's own state must never be the reason two sessions interfere. The run id the delegation skill generates names each delegation's artifacts: this skill names the isolated trees it creates by it, and handoff notes record each run directory's absolute path. Session-level evidence has a separate UUID, and links attempts to delegation run ids without replacing their ownership. This requirement is about the skill's state only. It does not ask the skill to make the work being done through it safe to run concurrently: whether two sessions can edit the same working tree at once is a property of that work and of the user's setup, not something the skill is responsible for detecting or preventing.
  • Invoking the skill activates it for the rest of the session, per "Staying active for the whole session" in SKILL.md; the two dependencies, by contrast, activate nothing when loaded.
  • The skill is meant to work on modern Linux and macOS. Commands, paths, and tools it prescribes must be available on both; nothing may rely on one without an equivalent for the other. No other platform is of concern, and the skill's text need not accommodate one.

Session evidence

Every activated session records its routing decisions and outcomes in a private UUID-named directory under ${XDG_STATE_HOME:-$HOME/.local/state}/scode-galaxy-brain/sessions/, outside the repository. It retains the same identity across tasks, compaction, and resume. Unknown prior identity is a reported gap, never permission to adopt another session's directory. Records persist until the user removes them; this skill does not garbage-collect them.

Versioned structured events cover considered units kept local as well as delegated units, exact requested and reported model/effort, route rationale and constraints, actual launches, continuations, verdicts, local fixes, escalations, reroutes, takeovers, and unfinished work. Linked unit, decision, attempt, and delegation run ids distinguish retries from new work. Other skills' process-defined spawns are recorded without imposing the delegation skill's gate on them. Pre-launch replacements explicitly link to the prior decision through supersedes_decision_id, closing it as superseded rather than unfinished or failed. Replacement does not complete the unit or erase the outcomes of launched attempts.

Usage is evidence-backed and optional: missing counts are null with a reason, not zero or estimates. Records preserve counter scope, cumulative-versus-incremental semantics, source, and model attribution so later statistics can avoid double-counting resumes, cached or reasoning subsets, and nested agents. Orchestrator usage is separate and may remain unavailable. Collection neither changes routing policy nor launches extra agents to obtain accounting. Recording failures are reported as evidence gaps without blocking otherwise authorized work or silently claiming complete data.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 03a9f7f. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated last month

README badge

README badge for scode/dotfiles/scode-galaxy-brain