Skills
skills/odysseyalive/claude-enforcer/skill-builder

skill-builder

SKILL.md

Skill Builder

Quick Commands

Command Action
/skill-builder Full audit: runs optimize + agents + hooks in display mode for all skills
/skill-builder audit Same as above
/skill-builder audit --quick Lightweight audit: frontmatter + line counts + priority fixes only
/skill-builder cascade [skill] Validation cascade analysis: detect over-validation suppressing output
/skill-builder dev [command] Run any command with skill-builder itself included

Directives

"When a decision needs to be made that isn't overtly obvious, and guesses are involved, AGENTS ARE MANDATORY, in order to provide additional input in decision making."

— Added 2026-02-22, source: user directive

"Each agent being created by this system always has to have an appropriate persona that is not being used anywhere else."

— Added 2026-02-22, source: user directive

"When deploying a Team, one of the team member's persona is a research assistant who will research the issue using read-only reference tools. Other team members may also make requests from the research assistant to help augment the outcome."

— Added 2026-02-23, source: user directive (tool specifics in references/agents-teams.md)

Commands

All commands operate in display mode by default. Add --execute to apply changes. Before executing any command, read its procedure file from references/procedures/.

Command Procedure Summary
audit audit.md Full system audit
audit --quick audit.md Lightweight: frontmatter + line counts
cascade [skill] cascade.md Validation cascade analysis (diagnostic only)
optimize [skill] optimize.md Restructure for context efficiency
optimize claude.md claude-md.md Extract domain content to skills
agents [skill] agents.md Analyze/create agents
hooks [skill] hooks.md Inventory/create hooks
new [name] new.md Create skill from template
inline [skill] [directive] inline.md Quick-add directive
skills skills.md List local skills
list [skill] list.md Show modes/options
verify verify.md Health check (headless-compatible)
ledger ledger.md Create Awareness Ledger
self-heal self-heal.md Install Self-Heal companion
update (inline below) Update to latest version

The update Command

Re-run the installer to update skill-builder to the latest version.

  1. Run the installer directly via Bash: bash -c "$(curl -fsSL https://raw.githubusercontent.com/odysseyalive/claude-enforcer/main/install)"
  2. Tell the user: "Restart Claude Code to load the updated skill." The current session still has the old skill loaded in memory, so start a new conversation. Once you're back, run /skill-builder audit — updates often add new recommendations that apply to your existing skills.

Self-Exclusion Rule

The skill-builder skill MUST be excluded from all actions (audit, optimize, agents, hooks, skills list) unless the command is prefixed with dev.

  • /skill-builder audit → audits all skills EXCEPT skill-builder
  • /skill-builder optimize some-skill → works normally
  • /skill-builder optimize skill-builder → REFUSED. Say: "skill-builder is excluded from its own actions. Use dev prefix: /skill-builder dev optimize skill-builder"
  • /skill-builder dev audit → includes skill-builder in the audit
  • /skill-builder dev optimize skill-builder → allowed

Detection: If the first argument after the command is dev, strip it and proceed with self-inclusion enabled. Otherwise, skip any skill whose name is skill-builder when iterating skills, and refuse if skill-builder is explicitly named as a target.

Self-heal exception: /skill-builder self-heal installs self-heal and is not subject to exclusion — this is installation, not analysis. However, self-heal integration embedded in other skills IS subject to self-exclusion — skill-builder does not embed self-heal awareness into itself.

Post-dev check: After any dev command that modifies skill-builder files, verify that the install script still covers all files. Glob skill-builder/**/*.md, compare against the files downloaded in the installer's loop, and flag any new/renamed/removed files that the installer doesn't handle. This prevents drift between the repo and what users receive on install.

Display/Execute Mode Convention

All sub-commands (optimize, agents, hooks) operate in two modes:

Mode Behavior Flag
Display (default) Read-only plan of what would change (none)
Execute Apply changes to files --execute

Rules

  1. Default is always display mode. Running /skill-builder optimize my-skill shows what would change without modifying anything.
  2. --execute triggers modifications. Running /skill-builder optimize my-skill --execute applies the changes.
  3. Audit always calls sub-commands in display mode, then offers the user a choice of which to execute.
  4. Execution requires a task plan. When --execute is invoked, the command MUST:
    • First produce a numbered task list using TaskCreate, one task per discrete action
    • Execute each task sequentially, marking progress via TaskUpdate
    • This ensures context can be refreshed mid-execution without losing track, no tasks get forgotten during long context windows, and the user can see progress and resume if interrupted
  5. Scope discipline during execution. Execute ONLY the tasks in the task list. Do not add bonus tasks, expand scope, or create deliverables not in the original plan. If execution reveals a new opportunity, note it in the completion report — do not act on it. The task list is the contract.
  6. Post-action chaining. Any action that modifies a skill (new, inline, adding directives) automatically chains into a scoped mini-audit for the affected skill — running optimize, agents, and hooks in display mode, then offering execution choices. Use --no-chain to suppress.

Core Principles

IMPORTANT: Never break anything.

Optimization is RESTRUCTURING, not REWRITING. The skill must behave identically after optimization.

YOU MUST:

  1. MOVE content, don't rewrite it — Relocate to new location, preserving wording. Exclude secrets, credentials, API keys, tokens, and passwords — these must never appear in output or be relocated.
  2. PRESERVE all directives exactly — User's words are sacred
  3. KEEP all workflows intact — Same steps, same order, same logic
  4. TEST nothing changes — After optimization, skill works identically

What optimization IS:

  • Moving reference tables to reference.md
  • Moving lookup tables and named references to reference.md
  • Adding grounding requirements
  • Creating enforcement hooks
  • Splitting into SKILL.md + reference.md

What optimization is NOT:

  • Rewriting instructions "for clarity"
  • Condensing workflows "for brevity"
  • Changing step order "for efficiency"
  • Removing "redundant" content
  • Summarizing user directives
  • Reorganizing workflow structure that enforces directives (see enforcement.md § "Behavior Preservation")

The test: If the original author reviewed the optimized skill, they should say "this does exactly what mine did, just organized differently."

Directives are sacred.

When a user says "Never use Uncategorized accounts," those exact words stay in the skill, unchanged, forever.

YOU MUST distinguish between:

Content Type Can Compress? Where It Lives
Directives (user's exact rules) NEVER Top of SKILL.md, unchanged
Reference (lookup tables, mappings, theory) YES Separate reference.md
Machinery (hooks, agents, chains) YES settings.json, hooks/, agents

The Sacred Directive Pattern

When a user gives you a rule, store it unchanged in a ## Directives section with exact wording, source, and date. Place at TOP of skill file. NEVER summarize or reword. Enforce with hooks when possible.

Grounding: Read references/templates.md § "SKILL.md Template" and references/procedures/directives.md for format and workflow.

§ Output Discipline — Cascade, Don't Scatter

Applies to ALL procedures universally. This is a behavioral constraint on how skill-builder presents its output.

Rules

  1. Own-skill actions (optimize, agents, hooks, ledger for skills managed by skill-builder): Always cascade into the current execution flow via AskUserQuestion + TaskCreate. Never present as standalone slash commands for manual invocation.
  2. Cross-skill actions (commands belonging to other skills like /awareness-ledger, /self-heal): Present in a clearly separated "Related Suggestions" footer, labeled as informational. Never mix into execution menus.
  3. Informational recommendations in conditional notes (e.g., "Run X to fix this"): Reframe as what the current procedure will do, or defer to the execution menu. Example: instead of "Run /skill-builder optimize to add eval protocol", say "Missing runtime eval protocol — flagged for optimization."
  4. Anti-pattern: Never end any procedure output with a list of slash commands the user must copy-paste and run manually. If an action is worth recommending, it's worth cascading.

Grounding

Before using any template, example, or pattern from reference material:

  1. Read the relevant file from references/
  2. State: "I will use [TEMPLATE/PATTERN] from references/[file] under [SECTION]"

Reference files:

Weekly Installs
10
Repository
odysseyalive/cl…enforcer
GitHub Stars
9
First Seen
Feb 26, 2026
Security Audits
Gen Agent Trust HubPass
SocketWarn
SnykWarn
Installed on
claude-code10
mcpjam9
kilo9
junie9
windsurf9
zencoder9