Claudify#

Guide users to convert functionality into the appropriate Claude Code automation type (Agent, Skill, Rule, Command, or Hook).

Topic Dispatch#

When this skill is invoked with a topic specifier (e.g., Skill("claudify", "improve") or /claudify improve), load and follow ONLY the matching topic file below — do NOT default to the creation-wizard body (Decision Matrix / Workflow) in this SKILL.md. A bare invocation (no topic, agentify, or create) uses the creation workflow below.

Topic argFilePurposecreate / agentify / (none)this SKILL.md (below)Convert functionality into an automation (Agent/Skill/Rule/Command/Hook)improveimprove.mdSelf-improving loop: retrospect + hook/skill review + pattern detectpersistpersist.mdKnowledge persistence: documentation + memory savebackground-pollingbackground-polling.mdScheduleWakeup/timeout polling discipline for 5min+ background dispatches

For improve / persist / background-polling, Read that topic file and follow it — the Decision Matrix and creation Workflow below apply only to the create/ agentify/(none) creation path. For any other/unrecognized topic arg, fall back to the create/ agentify/(none) creation workflow below rather than assuming a matching topic file exists.

Decision Matrix#

TypeWhen to useImplementationAgentHigh autonomy, multi-tool.claude/agents/name.mdSkillDomain expertise, logic.claude/skills/name/SKILL.mdRuleConstraints, styling.claude/rules/name.mdSlash CommandUser types /cmdSimple prompt templatesHookEvents (tool use, etc)Automation on actions

Full comparison: automation-decision-guide.md

Workflow#

Step 0: Output-size routing (HARD STOP — before Step 1)#

Before any large-output operation, decide: inline vs subagent vs script. Parent context bloat from claudify itself reduces the budget available for the actual automation creation work.

Routing decision matrix

OperationInline OK?Dispatch to subagent?Script?Read 1 resource file (template/guide)✅ Inline——Read 2+ large resource files (>200 lines each)❌✅ general-purpose subagent—Scan transcript for automation patterns (no target)❌✅ Explore or general-purpose—Enumerate ~/.claude/plugins/marketplaces/*/plugins/*/❌—✅ find / ls 1-linerCreate single agent/rule/command file (inputs known)✅ Inline——Create multi-topic skill (writer + multiple topic files)❌✅ skill-writer subagent—Marketplace remote search (WebFetch + parsing)❌✅ general-purpose subagent—

Don't / Do table

#Don'tDo1Read resources/agent-templates.md (358 lines) + automation-decision-guide.md + askuserquestion-patterns.md inline before creating one agentDispatch general-purpose subagent: "Read these 3 templates and create agent at <path> with name=X, tools=Y, description=Z. Return the created file path only."2Inline scan transcript for candidates by Read-ing the full session JSONLDispatch Explore subagent: "Find verbose tool-output patterns (>500 tokens repeated 2+ times) in conversation. Return candidate list (label + 1-line description) under 200 words."3Inline Glob + Read all marketplace plugin SKILL.md filesBash 1-liner: find ~/.claude/plugins/marketplaces/*/plugins/*/ -name SKILL.md -exec head -3 {} \; returns names without body4"Just one more Read" cumulative inline readingQuantify: if next operation expected to add >2K tokens to parent context AND result is not the deliverable itself, dispatch subagent5Dispatch subagent with vague prompt ("create the agent")Subagent prompt must include: target file path, name, tools, model, description (single-line YAML), trigger keywords. Return only the file path

Self-check (before each Read/Glob/WebFetch in claudify workflow)

1. Is the read result the deliverable, or intermediate context that the subagent's parent doesn't need?
2. Will this Read + subsequent Reads cumulatively add >2K tokens to parent context?
3. Is the operation parallelizable (multiple Reads, multiple files)?
4. If 1=intermediate OR 2=yes OR 3=yes → dispatch subagent

Subagent return contract: subagent returns only the deliverable path(s). No template dump, no intermediate analysis, no confirmation text.

Step 1: Identify Candidates#

If no target specified ("agentify" alone):

• Review conversation for automation candidates
• Look for: verbose outputs, multi-step workflows, repeated patterns
• MUST use multiSelect: true when presenting candidates (users often want multiple)

If target specified:

1. Check local marketplaces first:
   • ~/.claude/plugins/marketplaces/*/plugins/*/
   • Use /skill-dedup command to find overlaps
2. If not found locally, search remote: WebFetch https://claudemarketplaces.com/?search=[keyword]
3. If found, recommend existing or extend. If not, proceed to create

Step 1 Guards (HARD STOP):

• No immediate-execution options: AskUserQuestion options must all be automation types (Skill/Agent/Rule/Hook/Command). "Just execute now" or "Execute without automation" options are forbidden — claudify's purpose is creating/extending automation, not executing the underlying task.
• Scope check when extending existing skills: Before proposing to extend a global skill ( ~/.claude/skills/), verify the change is project-agnostic. Project-specific variables, hostnames, or environment values must NOT be added to global skills. If the automation is project-specific, recommend a project-level skill ( .claude/skills/) or a project rule ( .claude/rules/) instead.

Step 1.5: Merge logic / Grouping / AskUserQuestion#

When duplicate or similar automation candidates are found, confirm grouping and merge options with the user using AskUserQuestion.

Grouping criteria:

• Similarity in model or functionality
• Contextual similarity
• Keyword grouping (e.g., "openclaw management")

Merge criteria:

ConditionRecommendation3+ similar topicsmulti-topic SkillDifferent triggers + same skillGeneral AgentSimple instructions + Bash scriptSkill (Agent with skill)Complex multi-step + multiple toolsAgent

AskUserQuestion (merge options): "Candidates found. How should I structure them?" options:

• "Merge related topics into one multi-topic skill (e.g., openclaw: exec/gateway/test)"
• "Create separate agents for each functionality"
• "Skill + Agent combination (instruction=skill, implementation=agent)"

PROHIBITED: Do not create separate agents without merging candidates into a logical structure if they are related.

Step 2: Gather Requirements#

Use AskUserQuestion to clarify:

• Triggers: Commands or conditions that activate the automation
• Tools: Required tools or skills
• Scope: Global ( ~/.claude/) / Project ( .claude/)
• Language: code comments, variable names, documentation

Question patterns: askuserquestion-patterns.md

Step 3: Recommend Type#

Use the automation-decision-guide.md to recommend the best type.

Step 4: Create#

CRITICAL: Follow the creation method for each type. Apply Step 0 routing first — multi-template Reads or transcript scans dispatch to subagent.

Skill with scripts (If scripts are required, use skill directory structure):

text

skill-name/
├── SKILL.md          # frontmatter + documentation + node scripts/xxx.js (mode)
├── topic-a.md        # topic file
├── topic-b.md        # topic file
└── scripts/
    └── xxx.js        # actual logic (temp location, permanent storage)

• Do not create tmp_*.js in current directory; move to scripts/
• Call in SKILL.md: node <skill-dir>/scripts/xxx.js <mode>
• Use relative path from __dirname in scripts
• Separate implementation logic and topic files

Skill (no scripts): MUST use skill-writer (do NOT create directly)

text

Skill tool: skill: "project-automation:skill-writer"

Agent: Create in ~/.claude/agents/ or .claude/agents/

• agent-templates.md

Rules: Create in ~/.claude/rules/ or .claude/rules/

• rules-guide.md

Slash Command: Create in ~/.claude/commands/ or .claude/commands/

• slash-command-syntax.md

Hook: Add to settings.json

• hook-examples.md

Plugin (open source):

• plugin-creation.md

Step 5: Validate#

1. Register or copy to target location
2. Reload Claude Code:
   1. Manual sync to cache, OR
   2. New session to reload

No auto-sync hook currently exists — verified 2026-08-18: no plugin-cache-sync.sh is registered in settings.json/ settings.local.json, and no live copy of the script exists outside a Syncthing version-history backup. A marketplace checkout's installPath in ~/.claude/plugins/installed_plugins.json can point at a cache directory that either does not exist on disk (harness apparently falls back to reading the marketplace source directly in that case) or exists but has drifted stale relative to the marketplace checkout — see hook-kit/audit.md Step 3-D for the detection procedure. Until that gap is closed, manually verify ( diff the marketplace source against the cache installPath) after editing any plugin's hook scripts, rather than assuming this auto-sync exists.

Output Guidelines#

Keep responses concise:

1. List identified candidates (with multiSelect)
2. Summarize the recommended structure (Merge logic)
3. Provide the creation plan

ContextmultiSelectAutomation candidatestrue (users often want multiple)Merge options (Complexity)false (merge vs separation - mutually exclusive)Type selectionfalse (mutually exclusive)Scope selectionfalse (one location)Feature selectiontrue (additive choices)

Success Case#

Scenario (2026-03-09):

• Found 3 openclaw-related functions
• Proposed 3 options for merging
• Result: Implementation success, user satisfied

Key factors:

1. Identification of 3 functions
2. "Merge?" AskUserQuestion
3. merging skills using skill-writer (multi-topic)

Ralph Mode (AskUserQuestion bypass)#

Detection condition: do not judge by .ralph/ presence alone. Ralph mode requires all of:

1. .ralph/ directory exists AND
2. Environment variable RALPH_LOOP=1 is set (Ralph autonomous loop sets this)

Even when .ralph/ exists, an interactive user session is normal mode — AskUserQuestion works as usual.

Workflow Change:

StepUser InteractionWorkflowStep 1: Auto-detectAskUserQuestion (multiSelect)Summary info to .ralph/improvements.mdStep 1.5: Merge logic / Structure-improvements.md recordingStep 2: RequirementsAskUserQuestiontrigger/scope recommendation to improvements.mdStep 3: Type recommendationRecommend onlyimprovements.md recordingStep 4: ImplementationDirect actionPROHIBITED - Use [NEEDS_REVIEW] tagStep 5: ValidationValidationAuto validation (after changes are complete)

Scope note (HARD STOP): this table governs the creation workflow only (new agent/skill/rule/hook). It does not extend to passive, no-ask persistence steps elsewhere in this skill's topics ( improve/ persist RAG session/discovery-chunk store) — those follow the narrower carve-out in persist.md's own "Ralph Mode" section, not this table. Do not generalize "Ralph Mode = record-only, nothing runs" from this table alone.

improvements.md recording example

markdown

## Agentify Candidate (Implementation)

### [Candidate Name]
- **Context**: [Why it was found]
- **Recommended Type**: [Skill/Agent/Hook/Slash Command]
- **Recommended Structure**: [Topics]
- **Rationale**: [Why it's recommended]
- **Tag**: [NEEDS_REVIEW]

Self-Improvement#

After changes are complete, Self-improve based on conversation:

1. Identify failure and workaround patterns