lsp-understand
À propos
Cette compétence offre une exploration exhaustive du code en construisant une carte de dépendances navigable pour tout symbole ou fichier. Elle synthétise les relations entre les implémentations, les hiérarchies d'appels, les références et le code source pour aider les développeurs à comprendre des bases de code inconnues. Utilisez-la lorsque vous avez besoin d'une vue d'ensemble structurelle complète, au-delà d'une simple recherche de symbole.
Installation rapide
Claude Code
Recommandénpx skills add blackwell-systems/agent-lsp -a claude-code/plugin add https://github.com/blackwell-systems/agent-lspgit clone https://github.com/blackwell-systems/agent-lsp.git ~/.claude/skills/lsp-understandCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Requires the agent-lsp MCP server.
lsp-understand
Deep-dive exploration of unfamiliar code — given a symbol or file, synthesizes hover info, implementations, call hierarchy (bounded to 2 levels), all references, and source into a structured Code Map.
Read-only — does not modify any files.
Differentiation from lsp-explore
/lsp-explore is a single-symbol pass: given one symbol name, it runs hover +
implementations + call hierarchy + references and produces a per-symbol report.
Use lsp-explore for quick "what is this one thing" questions.
/lsp-understand is broader in three ways:
- Accepts a file path as input — explores all exported symbols in that file as a group (Mode B), rather than requiring a single symbol name.
- Synthesizes cross-symbol relationships — produces a dependency map showing how entry points call each other, share callers, or implement the same interface, rather than isolated per-symbol reports.
- Enforces a 2-level call hierarchy depth limit — prevents infinite recursion in deeply connected code.
Use lsp-understand for "how does this module work as a whole."
Input — Two Modes
Mode A (symbol): User provides a symbol name in dot notation
(e.g., "codec.Encode", "Handler.ServeHTTP").
Mode B (file): User provides an absolute file path. All exported symbols in the file become the entry points.
Prerequisites
Call mcp__lsp__get_server_capabilities before Step 2 to determine which
capabilities are available. Skip steps that require missing capabilities:
go_to_implementation: skip Step 2b ifimplementationProvider: falsefind_callers: skip Steps 2c and 2d ifcallHierarchyProvider: false; note in the Code Map output that call hierarchy was unavailable
Step 1 — Entry Point Resolution
Mode A: Single Symbol
Call mcp__lsp__go_to_symbol to locate the symbol definition:
mcp__lsp__go_to_symbol({
"symbol_path": "<dot-notation name>", // e.g. "codec.Encode"
"workspace_root": "<root>" // optional
})
→ returns: file_path, line, column (1-indexed)
Record file_path, line, and column. If go_to_symbol returns nothing,
report:
Symbol not found:
<name>Check the dot-notation path (e.g. "Package.Symbol") and ensure the workspace root covers the file.
Stop immediately — do not proceed to Step 2.
The single symbol becomes the sole entry point.
Mode B: File Path
Call mcp__lsp__open_document then mcp__lsp__list_symbols:
mcp__lsp__open_document({ "file_path": "<absolute path>" })
mcp__lsp__list_symbols({ "file_path": "<absolute path>" })
→ returns: list of symbols with kind, line, column
Filter to exported symbols:
- Go: uppercase first letter
- TypeScript/JavaScript:
exportkeyword - Rust:
pubvisibility
Cap at 10 exported symbols maximum. If more than 10 are found, prioritize top-level functions and types; skip constants and variables.
Each filtered symbol becomes an entry point with its file_path, line, and
column.
Step 2 — Per-Symbol Analysis
For each entry point, run the following sub-steps. Where possible, parallelize calls within each step.
2a — Type Info and Docs
Call mcp__lsp__inspect_symbol using position_pattern with the @@
marker (see references/patterns.md):
mcp__lsp__inspect_symbol({
"file_path": "<file>",
"position_pattern": "<symbol@@name>",
"line_scope_start": <line - 5>,
"line_scope_end": <line + 5>
})
→ returns: hover text with type signature and doc comment
Store result as hover_text. If the call fails or returns nothing, set
hover_text to an empty string. Do not stop.
2b — Implementations (capability-gated)
If implementationProvider is available in server capabilities:
mcp__lsp__go_to_implementation({
"file_path": "<file>",
"line": <line>,
"column": <column>
})
→ returns: list of concrete implementation locations
Skip if capability is absent. Record "not supported by this server" rather
than stopping.
2c — Incoming Call Hierarchy (bounded to 2 levels)
If callHierarchyProvider is available:
Level 1 — Direct callers:
mcp__lsp__find_callers({
"file_path": "<file>",
"line": <line>,
"column": <column>,
"direction": "incoming"
})
→ returns: list of direct caller functions with file and line
Level 2 — Callers of callers:
For each Level 1 caller, call mcp__lsp__find_callers once more:
mcp__lsp__find_callers({
"file_path": "<caller file>",
"line": <caller line>,
"column": <caller column>,
"direction": "incoming"
})
→ returns: Level 2 callers
STOP at Level 2 — do not recurse further under any circumstances.
If Level 2 callers > 10: summarize by count and file, do not list individually.
2d — Outgoing Calls (Level 1 only)
If callHierarchyProvider is available:
mcp__lsp__find_callers({
"file_path": "<file>",
"line": <line>,
"column": <column>,
"direction": "outgoing"
})
→ returns: list of functions this symbol calls
Level 1 only — no recursion.
2e — All References
mcp__lsp__find_references({
"file_path": "<file>",
"line": <line>,
"column": <column>,
"include_declaration": false
})
→ returns: every usage site across the workspace
Group by file and count distinct files.
2f — Source
mcp__lsp__get_symbol_source({
"file_path": "<file>",
"line": <line>,
"column": <column>
})
→ returns: implementation body
Step 3 — Synthesize Relationships
After analyzing all entry points, identify cross-symbol relationships:
- Internal calls: Which entry points call each other? (from outgoing calls in Step 2d)
- Shared callers: Which entry points are called by the same Level 1 callers?
- Shared interface: Which entry points implement the same interface? (from Step 2b)
This synthesis step is what distinguishes /lsp-understand from running
multiple /lsp-explore calls. The output is a dependency map, not isolated
per-symbol reports.
Step 4 — Output: Code Map
Produce a structured Code Map with these sections:
## Code Map: <target>
### Summary
<2-3 sentence description of what this code does, synthesized from
hover docs and source reading>
### Symbols (<N> analyzed)
#### <SymbolName>
- **Type:** <type signature from hover>
- **Source:** <file:line>
- **Incoming callers (L1):** <list; count only if > 5>
- **Incoming callers (L2):** <summarized; e.g., "called by 3 HTTP handlers">
- **Outgoing calls:** <what this symbol calls>
- **Implements:** <interface name, if applicable>
- **References:** N sites across M files
### Dependency Relationships
<symbols that call each other, as a simple text diagram or list>
e.g.:
HandlerA → Parse → Validate
HandlerB → Parse
### Entry Points to This Code
<top-level callers that are NOT in this file — where does outside code
call in?>
### Depth-limit Note
Call hierarchy stopped at 2 levels. <N> additional callers exist beyond
Level 2 — use /lsp-explore on specific symbols to drill deeper.
Depth Control Rules
These limits are hard constraints — never exceed them:
- Incoming call hierarchy recursion stops at Level 2
- Outgoing calls: Level 1 only, no recursion
- If Level 2 callers > 10: summarize by count and file, do not list individually
- Do NOT follow call chains beyond these limits under any circumstances
Example
Goal: understand how the file pkg/codec/encoder.go works as a whole
Step 1 — Mode B (file path)
open_document: pkg/codec/encoder.go
list_symbols: pkg/codec/encoder.go
→ exported symbols: Encoder (type), Encode (func), Reset (func), NewEncoder (func)
→ 4 exported symbols (under 10 cap)
get_server_capabilities
→ go_to_implementation: supported
→ find_callers: supported
Step 2 — Per-symbol analysis (run in parallel across symbols)
Symbol: NewEncoder (pkg/codec/encoder.go:12)
inspect_symbol → "func NewEncoder(w io.Writer) *Encoder"
go_to_implementation → 0 (concrete function)
find_callers incoming L1 → 5 callers
find_callers incoming L2 → 3 callers of those callers
find_callers outgoing → calls: bufio.NewWriter
find_references → 5 sites in 3 files
get_symbol_source → implementation body
Symbol: Encode (pkg/codec/encoder.go:28)
inspect_symbol → "func (e *Encoder) Encode(v any) error"
go_to_implementation → implements codec.Encoder interface
find_callers incoming L1 → 8 callers (listed)
find_callers incoming L2 → > 10: "12 additional callers across 5 files"
find_callers outgoing → calls: NewEncoder, e.w.Flush
find_references → 8 sites in 5 files
get_symbol_source → implementation body
(similar for Encoder type and Reset func...)
Step 3 — Synthesize relationships
- Encode calls NewEncoder (internal dependency)
- NewEncoder and Encode share callers in cmd/main.go
- Encode implements codec.Encoder interface
## Code Map: pkg/codec/encoder.go
### Summary
This file implements a streaming JSON encoder backed by a buffered writer.
NewEncoder constructs an Encoder wrapping any io.Writer; Encode serializes
values and flushes. Reset allows reuse without allocation.
### Symbols (4 analyzed)
#### NewEncoder
- **Type:** func NewEncoder(w io.Writer) *Encoder
- **Source:** pkg/codec/encoder.go:12
- **Incoming callers (L1):** cmd.main, app.Start, loader.Load, test.Setup, bench.Run
- **Incoming callers (L2):** 3 callers across 2 files
- **Outgoing calls:** bufio.NewWriter
- **Implements:** n/a
- **References:** 5 sites across 3 files
#### Encode
- **Type:** func (e *Encoder) Encode(v any) error
- **Source:** pkg/codec/encoder.go:28
- **Incoming callers (L1):** 8 callers (cmd/main.go, internal/app.go, ...)
- **Incoming callers (L2):** 12 additional callers across 5 files (depth limit reached)
- **Outgoing calls:** NewEncoder, e.w.Flush
- **Implements:** codec.Encoder
- **References:** 8 sites across 5 files
...
### Dependency Relationships
cmd.main → NewEncoder → bufio.NewWriter
cmd.main → Encode → NewEncoder
Encode → Reset
### Entry Points to This Code
- cmd.main (cmd/main.go:14)
- app.Start (internal/app.go:31)
- loader.Load (internal/loader.go:55)
### Depth-limit Note
Call hierarchy stopped at 2 levels. 12 additional callers exist beyond
Level 2 for Encode — use /lsp-explore on specific symbols to drill deeper.
Dépôt GitHub
Questions fréquentes
Qu’est-ce que le Skill lsp-understand ?
lsp-understand est un Skill Claude créé par blackwell-systems. Un Skill regroupe des instructions et des ressources que Claude charge à la demande pour effectuer des tâches liées à lsp-understand sans consigne supplémentaire.
Comment installer lsp-understand ?
Utilisez les commandes d’installation de cette page : ajoutez lsp-understand à Claude Code comme plugin ou clonez son dépôt dans votre dossier skills, puis redémarrez Claude pour charger le Skill.
À quelle catégorie appartient lsp-understand ?
lsp-understand appartient à la catégorie Méta.
lsp-understand est-il gratuit ?
Oui. lsp-understand est référencé sur AIMCP et son installation est gratuite.
Compétences associées
Cette compétence propose une configuration éprouvée en production pour Content Collections, un outil axé sur TypeScript qui transforme des fichiers Markdown/MDX en collections de données typées de manière sûre avec une validation Zod. Utilisez-la lors de la création de blogs, de sites de documentation ou d'applications Vite + React riches en contenu pour garantir la sécurité de typage et la validation automatique du contenu. Elle couvre tout, de la configuration du plugin Vite et de la compilation MDX à l'optimisation des déploiements et la validation des schémas.
Cette compétence permet aux développeurs de créer des applications avec la plateforme de marchés prédictifs Polymarket, incluant l'intégration d'API pour le trading et les données de marché. Elle fournit également une diffusion de données en temps réel via WebSocket pour surveiller les transactions en direct et l'activité du marché. Utilisez-la pour mettre en œuvre des stratégies de trading ou pour créer des outils traitant les mises à jour de marché en direct.
Cette compétence aide les développeurs à créer des plugins OpenCode qui s'interconnectent avec plus de 25 types d'événements tels que les commandes, les fichiers et les opérations LSP. Elle fournit la structure du plugin, les spécifications de l'API événementielle et les modèles d'implémentation pour les modules JavaScript/TypeScript. Utilisez-la lorsque vous avez besoin d'intercepter, de surveiller ou d'étendre le cycle de vie de l'assistant IA OpenCode avec une logique personnalisée pilotée par les événements.
SGLang est un framework de service LLM haute performance spécialisé dans la génération rapide et structurée pour les workflows JSON, regex et agentiques grâce à son cache de préfixe RadixAttention. Il offre une inférence nettement plus rapide, particulièrement pour les tâches avec des préfixes répétés, ce qui le rend idéal pour les sorties complexes et structurées ainsi que les conversations multi-tours. Choisissez SGLang plutôt que des alternatives comme vLLM lorsque vous avez besoin d'un décodage contraint ou que vous construisez des applications avec un partage étendu de préfixes.
