Sk claude code conventions
Skill gustavo-meilus/superpipelines/skills/sk-claude-code-conventions
Provides canonical engineering standards for authoring agents, skills, hooks, and pipeline orchestration in Claude Code. Use when authoring or modifying agents, skills, hooks, or pipeline-orchestration artifacts — covers model-tier selection, prompt-cache discipline, frontmatter schemas, and progressive-disclosure rules.From its SKILL.md
npx -y skills add gustavo-meilus/superpipelines --skill sk-claude-code-conventionsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.8 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it
Claude Code Conventions — Reference
<glossary> <term name="Adaptive Thinking">Model-driven reasoning scaled by the `effort` parameter.</term> <term name="Prompt Caching">A mechanism to persist static context (system prompts, skills) to reduce latency and cost.</term> <term name="Progressive Disclosure">Structuring content into layers (Body → References) to minimize context bloat.</term> </glossary>Model & Reasoning Selection
<model_tier_table>
| Tier | Model ID | Application |
|---|---|---|
triage | claude-haiku-4-5-20251001 | High-volume, low-latency classification. |
fast | claude-haiku-4-5-20251001 | Reviews and bounded worker steps. |
medium | claude-sonnet-4-6 | Default for pipelines, agents, and skills. |
deep | claude-opus-4-8 | High-complexity architecture or auditing tasks. |
| </model_tier_table> |
Caching & Context Discipline
<protocol> ### 1. CACHE STABILITY - **Static First**: Context ordered as System Prompt → Skills → Tools → Conversation History. - **Invariant Toolsets**: Tool and skill lists must not be mutated mid-session; mutation invalidates the entire prefix. - **Static Headers**: Dynamic timestamps or mutable data must not appear in the static system prefix.2. CACHE BREAKPOINTS
- Up to 4
cache_controlbreakpoints are supported. - The final breakpoint belongs on the last block that remains identical across requests.
- Agentic loops use 1-hour TTL to prevent cache expiration between turns. </protocol>
Artifact Schemas
Skill Frontmatter
<invariant> `SKILL.md` bodies must not exceed 500 lines. Descriptions must use third-person voice and focus strictly on triggering conditions. </invariant>name: {lowercase-hyphens}
description: {triggering-only, third person}
effort: {low | medium | high | xhigh | max}
disable-model-invocation: {boolean}
user-invocable: {boolean}
Agent Frontmatter
<invariant> Agent bodies must not exceed 150 lines. Depth belongs in companion `<agent>-references/references/*.md` files. </invariant>name: {identifier}
description: {action-oriented trigger}
effort: {high for architects, medium for workers}
isolation: worktree
skills: [sk-4d-method, sk-spec-driven-development]
<invariant>
Plugin-scope enforcement rule: Claude Code ignores `permissionMode`, `hooks`, and `mcpServers` on subagents loaded from a plugin — only `tools:` and `disallowedTools:` are enforced on that path. Reviewer write-deny for plugin-shipped agents therefore rests entirely on the tool allowlist; `permissionMode` in `agents/*.md` is declarative intent, not a barrier. Project-scope agents (including agents MATERIALIZE writes under `.claude/agents/superpipelines/{P}/`) enforce `permissionMode` normally. No document may claim `permissionMode` enforcement for plugin-shipped agents (auditor BUNDLE-08).
</invariant>
Progressive Disclosure Checklist
<checklist> - [ ] `SKILL.md` body ≤500 lines. - [ ] References are exactly one level deep from `SKILL.md`. - [ ] References >100 lines include a Table of Contents. - [ ] Descriptions are third-person and triggering-only. - [ ] Examples are concrete and specific to the domain. </checklist> <invariants> - `memory: project` must not appear in agent frontmatter. - All relative paths must be resolved via `${CLAUDE_PLUGIN_ROOT}` or `${CLAUDE_PLUGIN_DATA}`. - All pipeline state must be persisted to the canonical `pipeline-state.json` location. </invariants>Reference Files
sk-pipeline-state/SKILL.md— State management rules.sk-pipeline-paths/SKILL.md— Scope-aware path resolution.sk-4d-method/SKILL.md— Per-invocation wrapper.sk-spec-driven-development/SKILL.md— SDD protocol.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most context ai engineering skills give in ~1.1k tokens
Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06
- Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
- Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
- Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
- Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
- Use the least powerful model capable of the taskin 33 of 1328, across 26 files
- Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
- Perform a task review after each implementationin 31 of 1328, across 24 files
- Extract all tasks and context from the planin 29 of 1328, across 20 files
- Provide full task text to subagentsin 28 of 1328, across 20 files
- Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
- Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
- Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files
Said here and by no other author read
- use tier-based model resolution for pipelines
- order context as system prompt then skills then tools
- keep tool and skill lists immutable mid-session
- place cache breakpoints on the final static block
- limit skill bodies to 500 lines
- limit agent bodies to 150 lines
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.