blooming-blockery-cli（blooming blockery CLI 搜索）

blooming-blockery-cli

Use this skill when operating Blooming Blockery CLI (blooming-blockery) and when shaping content into tree structure for better LLM context quality.

Primary goal: treat Blooming Blockery as a tree-structure based design documentation editor. Organize information into a logically correct tree so each block carries one coherent idea, and unrelated material is split into separate branches.

This skill is not only command syntax reference. It is a structure-and-context guide for deciding where information should live in the tree.

Motivation: From Vague Ideas to Implementation

Use the tree to move from uncertainty to precision:

1. Start with broad project ideas or high-level concepts.
2. Break ideas into smaller design decisions and scope boundaries.
3. Refine decisions into explicit specifications, such as requires, ensures, and invariants.
4. Continue refining specification blocks into implementable units.
5. Connect implementation steps back to the exact specification branch they satisfy.

Agents using this skill should preserve this top-down refinement flow whenever possible.

When to use

• The user asks how to use Blooming Blockery CLI (blooming-blockery) from terminal scripts.
• The user needs to restructure mixed or overloaded notes into a cleaner tree.
• The user needs LLM-ready context around a specific block.
• The user hits CLI parsing errors with block IDs, mount formats, or panel states.
• The user needs copyable command patterns for repeated authoring workflows.

Mental Model: Context Envelope Per Block

From src/llm/context.rs, one block's LLM context is a BlockContext made of:

• lineage: ordered chain from root to target, where the last item is the target block itself.
• existing_children: direct child point texts only.
• friend_blocks: cross-references with optional perspective, optional friend lineage telescope, and optional friend children telescope.

Implications for agents using this skill:

• Keep each block focused. If one block mixes unrelated concerns, split it.
• Prefer children for decomposition of one topic.
• Prefer siblings or new roots for unrelated topics.
• Prefer friends for cross-cutting links that are related but not hierarchical.
• Add --perspective when a friend relation is ambiguous.
• Use telescope flags intentionally to expand friend context only when needed.

Context-Hygiene Rules for Tree Editing

1. One block, one intent.
2. If two statements cannot share the same immediate design question, split them.
3. Keep parent-child logic strict: each child should refine, justify, or execute the parent.
4. Keep implementation detail under the relevant design/spec parent.
5. Do not bury unrelated decisions deep in another topic's lineage.
6. Preserve refinement order: concept -> spec (requires/ensures/invariants) -> implementation.
7. Use blooming-blockery context <BLOCK_ID> to inspect context before LLM draft commands.
8. Use blooming-blockery nav lineage <BLOCK_ID> when checking whether placement is right.
9. Use blooming-blockery friend add instead of forced tree nesting for cross-topic links.

When Tree Structure Is Hard, Use Friend Blocks

Some content is inherently cross-cutting and does not fit cleanly under one single parent branch. In those cases:

• Keep the main tree logically clean.
• Use blooming-blockery friend add <TARGET_ID> <FRIEND_ID> for remote references.
• Add --perspective to explain why the remote block matters to the target.
• Enable telescope flags only when extra friend lineage/children are truly needed.

Friend blocks are the escape hatch for useful context that should not distort the core tree semantics.

CLI Invocation Convention

Blooming Blockery is a single binary: run without subcommands to launch the GUI, or with subcommands for CLI automation. All examples below use the full form:

blooming-blockery ...

Use blooming-blockery <subcommand>... in instructions and automation scripts.

Global Flags

All commands support these global flags:

• --store <PATH>: Path to the block store file (defaults to app data path)
• --verbose: Enable verbose output (currently reserved)
• --output <FORMAT>: Output format - table (default) or json

Block ID Format

Block IDs use a clean format like 1v1, 2v3 where:

• First number = slot index in the store
• v = separator
• Second number = generation counter (increments on reuse)

Batch-capable commands also accept comma-separated IDs in a single ID argument:

blooming-blockery show 1v1,2v1,3v1
blooming-blockery tree add-child 1v1,2v1 "Shared child text"

Batch execution is continue-on-error: all targets are attempted and errors are reported after processing completes.

Context-First Workflow Patterns

1) Split unrelated information into separate branches

# Add a sibling when topic shifts at same hierarchy level
blooming-blockery tree add-sibling <BLOCK_ID> "New topic"

# Wrap current block when introducing a new organizing parent
blooming-blockery tree wrap <BLOCK_ID> "Parent concept"

# Move misplaced content under the correct parent
blooming-blockery tree move <SOURCE_ID> <TARGET_ID> --under

2) Inspect context before asking LLM features

blooming-blockery nav lineage <BLOCK_ID>
blooming-blockery context <BLOCK_ID> --output json
blooming-blockery friend list <BLOCK_ID> --output json

3) Keep cross-cutting references explicit

blooming-blockery friend add <TARGET_ID> <FRIEND_ID> \
  --perspective "reference architecture constraint" \
  --telescope-lineage

Use --telescope-lineage and --telescope-children only when the added friend scope is needed for the specific LLM task.

Command Reference

Query Commands

# List all root block IDs
blooming-blockery roots
blooming-blockery roots --output json

# Show block details
blooming-blockery show <BLOCK_ID>
blooming-blockery show 1v1 --output json

# Search blocks by text (case-insensitive substring)
blooming-blockery find "search query"
blooming-blockery find "TODO" --limit 10

# Edit the text content of a block
blooming-blockery point <BLOCK_ID> "New text content"
blooming-blockery point 1v1 "Updated text"

Tree Structure Commands

# Add child block under parent (parent must not be a mount)
blooming-blockery tree add-child <PARENT_ID> "Text content"
blooming-blockery tree add-child 1v1 "My new idea"

# Add sibling after a block
blooming-blockery tree add-sibling <BLOCK_ID> "Text content"
blooming-blockery tree add-sibling 1v1 "Next sibling"

# Wrap a block with a new parent
blooming-blockery tree wrap <BLOCK_ID> "Parent text"
blooming-blockery tree wrap 1v1 "New parent section"

# Duplicate a subtree
blooming-blockery tree duplicate <BLOCK_ID>
blooming-blockery tree duplicate 1v1

# Delete a subtree (removes block and all descendants)
blooming-blockery tree delete <BLOCK_ID>
blooming-blockery tree delete 1v1

# Move block relative to target
blooming-blockery tree move <SOURCE_ID> <TARGET_ID> --before
blooming-blockery tree move <SOURCE_ID> <TARGET_ID> --after
blooming-blockery tree move <SOURCE_ID> <TARGET_ID> --under

Navigation Commands

# Get next visible block in DFS order
blooming-blockery nav next <BLOCK_ID>
blooming-blockery nav next 1v1

# Get previous visible block
blooming-blockery nav prev <BLOCK_ID>
blooming-blockery nav prev 2v1

# Get lineage (ancestor chain)
blooming-blockery nav lineage <BLOCK_ID>
blooming-blockery nav lineage 1v1

# Search-aware navigation
blooming-blockery nav find-next <BLOCK_ID> "query"
blooming-blockery nav find-prev <BLOCK_ID> "query" --no-wrap

Draft Commands (LLM suggestions)

# Set amplification draft (rewrite + proposed children)
blooming-blockery draft amplify <BLOCK_ID> --rewrite "Refined text" --children "Child 1" "Child 2"
blooming-blockery draft amplify 1v1 --children "Just kids"

# Set distillation draft (condensed version)
blooming-blockery draft distill <BLOCK_ID> --reduction "Condensed text" --redundant-children 2v1 3v1

# Set instruction draft (user-authored LLM instructions)
blooming-blockery draft instruction <BLOCK_ID> --text "Make this more concise"

# Set probe draft (LLM response to ask query)
blooming-blockery draft probe <BLOCK_ID> --response "The key insight is..."

# List all drafts for a block
blooming-blockery draft list <BLOCK_ID>

# Clear drafts (use --all or specific flags)
blooming-blockery draft clear <BLOCK_ID> --all
blooming-blockery draft clear <BLOCK_ID> --amplify
blooming-blockery draft clear <BLOCK_ID> --distill --instruction

Fold (Collapse) Commands

# Toggle fold state
blooming-blockery fold toggle <BLOCK_ID>

# Get fold status
blooming-blockery fold status <BLOCK_ID>

Friend (Cross-reference) Commands

# Add friend block with optional perspective
blooming-blockery friend add <TARGET_ID> <FRIEND_ID> --perspective "Related design"
blooming-blockery friend add <TARGET_ID> <FRIEND_ID> --telescope-lineage --telescope-children

# Remove friend
blooming-blockery friend remove <TARGET_ID> <FRIEND_ID>

# List friends
blooming-blockery friend list <TARGET_ID>

Mount (External File) Commands

# Set mount path (block must be leaf, no children)
blooming-blockery mount set <BLOCK_ID> <PATH> [--format json|markdown]
blooming-blockery mount set 1v1 /data/external.json
blooming-blockery mount set 1v1 /notes/notes.md --format markdown

# Expand mount (load external file)
blooming-blockery mount expand <BLOCK_ID>

# Collapse mount (remove loaded blocks, restore mount node)
blooming-blockery mount collapse <BLOCK_ID>

# Move mount backing file and update metadata
blooming-blockery mount move <BLOCK_ID> <PATH>

# Inline mounted content into current store
blooming-blockery mount inline <BLOCK_ID>

# Inline all mounts recursively under a subtree
blooming-blockery mount inline-recursive <BLOCK_ID>

# Extract subtree to external file
blooming-blockery mount extract <BLOCK_ID> --output <PATH> [--format json|markdown]
blooming-blockery mount extract 1v1 --output /backup/notes.json

# Persist all expanded mounts to their source files
blooming-blockery mount save

# Show mount info
blooming-blockery mount info <BLOCK_ID>

Panel Commands

# Set block panel state
blooming-blockery panel set <BLOCK_ID> friends
blooming-blockery panel set <BLOCK_ID> instruction

# Get block panel state
blooming-blockery panel get <BLOCK_ID>

# Clear block panel state
blooming-blockery panel clear <BLOCK_ID>

Context Command

# Get LLM context for a block (lineage, children, friends)
blooming-blockery context <BLOCK_ID>
blooming-blockery context 1v1

Common Error Patterns

• UnknownBlock: Block ID not found in store
• InvalidOperation:

- Parent is a mount (cannot add children) - Source is ancestor of target (cycle in move) - Block has children (cannot set mount) - Attempting to add self as friend

• IoError: Failed to read/write mount file

Agent Guidance: Choosing Structure Intentionally

When helping users author specs/docs, prefer these decisions:

• If content answers different design questions, split into sibling branches.
• If content elaborates one parent claim, place it as children.
• Keep the logical chain visible: concept -> spec -> implementation.
• Encourage explicit spec nodes with requires, ensures, and invariants before implementation-heavy branches.
• If two branches should remain separate but mutually relevant, link via friends.
• If friend meaning is unclear, add --perspective immediately.
• Before generating drafts, inspect with blooming-blockery context and ensure lineage points to the intended scope.

Practical Tips

1. Use --output json for scripting and parsing results
2. Block IDs are case-insensitive
3. Mount format defaults to json but supports markdown

- mount extract --format overrides path-extension inference

1. Batch-capable commands support comma-separated IDs:

- show, point - tree add-child|add-sibling|wrap|duplicate|delete|move - nav next|prev|lineage - friend add|remove - fold toggle|status - draft instruction|probe|list|clear - mount set|expand|collapse|move|extract|inline|inline-recursive|info - context - For mount set|move|extract batch mode, pass a directory-like path (existing/dir or a path without extension), and each target will use <BLOCK_ID>.<ext> under that directory.

1. Panel states are friends or instruction
2. Run blooming-blockery without subcommands for GUI; with subcommands for CLI
3. If users report low-quality LLM responses, first fix tree placement and friend/telescope scope before rewriting prompts