omcs
Use when a substantial engineering task needs OMCS risk-gated orchestration, including when the user says use OMCS to solve this issue.
Skill by Rafcin
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/rafcin-omcs
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/omcs
What's inside
- Skill name
omcs- Description
- Use when a substantial engineering task needs OMCS risk-gated orchestration, including when the user says use OMCS to solve this issue.
- License
- MIT
- Instructions
- 118 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# OMCS OMCS is a thin routing kernel for one evidence-led delivery run. The primary context owns intent, routing, integration, verification, and acceptance. Orchestration must remove a concrete bottleneck or reduce a material risk. ## Entry conditions Use an effective profile already supplied by the runtime or current request. Otherwise apply the built-in `auto` defaults in memory. Only an explicit user choice selects `fast`, `thorough`, or `council`; urgency language is not a profile override. Do not require a CLI configuration command to begin work. Classify only observable signals: - `settled` means requirements, owned scope, interfaces, and acceptance oracle are sufficient for bounded execution. - Consequence is separate from uncertainty. Consequence is `material` for security, credentials, irreversible or external state, persistent data, dependencies, architecture, or compatibility with meaningful downstream impact. A narrow public or user-visible surface is not material by default. - Uncertainty is `material` when requirements, root cause, interfaces, or acceptance evidence are weak. Otherwise it is `low`. - Blast radius is `narrow`, `moderate`, or `wide` from the actual changed surfaces and downstream consumers. - `delegationValue` is true only when a complete bounded packet removes a real critical-path bottleneck or needs a specialist capability. Being technically delegable is not enough. Apply the smallest sufficient route: - `auto` and `fast` default to `solo`. - Require fresh review in `auto` or `fast` only when material consequence combines with material uncertainty or wide blast radius. - A one-auxiliary review route is `audit`: the primary implements and a fresh `omcs_reviewer` audits. - Select `delegate` only for settled, low-uncertainty work with concrete delegation value. Choose visual work → `omcs_designer`, wide non-visual work → `omcs_terra_fixer`, otherwise → `omcs_fixer`. - `thorough` and `council` always retain fresh review. They use `full` only when the same delegation-value test passes; otherwise they use `audit`. - `auto` and `fast` use at most one auxiliary. `thorough` may use one implementer plus one reviewer. Council advisers are explicit, read-only, and do not implement. After selecting an auxiliary candidate, preflight only the selected auxiliary. If an optional implementer or supporting specialist is unavailable, reroute narrow work visibly to `solo` and do not try another lane. If required fresh review is unavailable, fail closed with the missing capability. Council also fails closed unless two distinct advisory lanes are proven from non-secret metadata. Declare the result before mutation or delegation; a minimal read-only inspection needed to classify the task is allowed: ```text OMCS ROUTE profile: auto | fast | thorough | council mode: solo | delegate | audit | full risk: <consequence; uncertainty; blast radius> skills: <triggered skills> agents: <primary and selected auxiliaries> budget: <auxiliary limit; one final verification path; stop after green> approval: material-decisions ``` ## Scope limit Load focused skills only after a named trigger: - material ambiguity → `context` or `deep-interview`; - repository-mapping bottleneck → `codemap` or `omcs_explorer`; - current external behavior → `research` or `omcs_librarian`; - difficult reproducible diagnosis → `diagnose` or `omcs_oracle`; - material interface decision → `codebase-design`; - multi-step, delegated, persistent, or risky work → `plan`; - observable behavior change or regression → `tdd`; - concrete changed-file smell → `ai-slop-cleaner`; - justified behavior-preserving reduction → `simplify`; - final acceptance proof → `verification`; and - a review route → `code-review`. An auxiliary substitutes for the corresponding primary-context work; do not repeat its mapping, research, implementation, or review. Inspect its evidence only at the integration or acceptance boundary. Every delegation packet states objective, exact ownership, interfaces, exclusions, required evidence, and concurrent-work preservation. The primary never accepts a worker summary in place of inspecting the actual result. Pause only for a material decision that changes accepted scope, persistent or public contracts, security or credential ownership, architecture, dependencies, or irreversible external state. OpenCodex owns external transport and provider credentials. ## Exit evidence Use one final verification path proportionate to the acceptance claim. Repeat a command only after relevant inputs changed or when its prior evidence was incomplete. Run cleanup only when a concrete named finding exists in changed files; any cleanup edit invalidates affected verification and review. Acceptance evidence is a binding stop condition. Once the requested behavior, tests, and required review are green, stop. Do not make a post-green edit unless a named unresolved finding remains; if one is necessary, rerun affected verification and fresh review. Receipts may record only profile, route, triggered skills, selected agents, approval, command outcomes, and review verdict. Never record prompts, source text, paths, provider metadata, raw output, or secrets. ## Next gate Return an outcome-first acceptance report with the implemented result, fresh evidence, remaining limitation, and the explicit stop condition. A review returns `ship`, `fix-first`, or `rethink`; it never implies merge, deployment, or external-write approval.
From Rafcin/oh-my-codex-slim, MIT.
More from Rafcin/oh-my-codex-slim
- 0.526score
codemap
Rafcin
Build or refresh a concise repository map when a codebase's modules, entry points, and dependency boundaries must be understood.
SkillBFreshness: BClaude Code +9 - 0.526score
deepwork
Rafcin
Execute an already-approved, multi-step implementation plan with sustained focus, explicit ownership, and continuous verification.
SkillBFreshness: BClaude Code +9 - 0.524score
codebase-design
Rafcin
Use when a module boundary, interface, domain model, or architecture change needs a durable design decision before implementation.
SkillBFreshness: BClaude Code +9 - 0.524score
tdd
Rafcin
Implement a feature or bug fix test-first when the user asks for red-green development or the change needs a regression seam.
SkillBFreshness: BClaude Code +9 - 0.523score
omcs-orchestrate
Rafcin
Use when an existing workflow invokes the legacy OMCS orchestration skill name for a substantial engineering task.
SkillBFreshness: BClaude Code +9 - 0.523score
context
Rafcin
Use when a material product or engineering decision remains unclear after repository evidence has been inspected.
SkillBFreshness: BClaude Code +9 - 0.523score
implement
Rafcin
Use when a bounded implementation task has settled requirements, declared ownership, and a credible behavior seam.
SkillBFreshness: BClaude Code +9 - 0.521score
verification
Rafcin
Prove a completed change against explicit acceptance criteria before claiming completion, release readiness, or runtime success.
SkillBFreshness: BClaude Code +9 - 0.521score
deep-interview
Rafcin
Clarify an underspecified product or engineering request through one focused question at a time before planning or implementation.
SkillBFreshness: BClaude Code +9 - 0.521score
diagnose
Rafcin
Diagnose a reproducible bug or performance regression by building a tight feedback loop and testing ranked hypotheses before proposing a fix.
SkillBFreshness: BClaude Code +9 - 0.521score
plan
Rafcin
Turn settled requirements into an implementation plan with exact files, interfaces, tests, verification, and safe commit boundaries.
SkillBFreshness: BClaude Code +9 - 0.521score
simplify
Rafcin
Reduce accidental complexity in working code when behavior must remain unchanged and the user wants a smaller, clearer implementation.
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Aug 28, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/Rafcin/oh-my-codex-slim
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from Rafcin
- 0.526score
codemap
Rafcin
Build or refresh a concise repository map when a codebase's modules, entry points, and dependency boundaries must be understood.
SkillBFreshness: BClaude Code +9 - 0.526score
deepwork
Rafcin
Execute an already-approved, multi-step implementation plan with sustained focus, explicit ownership, and continuous verification.
SkillBFreshness: BClaude Code +9 - 0.524score
codebase-design
Rafcin
Use when a module boundary, interface, domain model, or architecture change needs a durable design decision before implementation.
SkillBFreshness: BClaude Code +9 - 0.524score
tdd
Rafcin
Implement a feature or bug fix test-first when the user asks for red-green development or the change needs a regression seam.
SkillBFreshness: BClaude Code +9 - 0.523score
context
Rafcin
Use when a material product or engineering decision remains unclear after repository evidence has been inspected.
SkillBFreshness: BClaude Code +9 - 0.523score
implement
Rafcin
Use when a bounded implementation task has settled requirements, declared ownership, and a credible behavior seam.
SkillBFreshness: BClaude Code +9
Cite this page
omcs skill. Agent Mart. https://agentmart.dev/skills/rafcin-omcs (last updated 2026-10-24)