lsp-refactor
About
The lsp-refactor skill provides an end-to-end safe refactoring workflow that coordinates blast-radius analysis, speculative previews, build verification, and affected test execution. It combines multiple LSP operations into a single sequence to ensure changes are safe before applying them. Use this when you need to confidently refactor code with automated safety checks and test correlation.
Quick Install
Claude Code
Recommendednpx 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-refactorCopy and paste this command in Claude Code to install this skill
Documentation
Requires the agent-lsp MCP server.
lsp-refactor
End-to-end safe refactor workflow. Sequences blast-radius analysis, speculative preview, disk apply, build verification, and targeted test execution in one coordinated pass.
This skill does NOT replace lsp-safe-edit or lsp-impact.
lsp-safe-editwraps a single edit with before/after diagnostic comparison — use it when you need to make one targeted change with careful error diffing.lsp-impactis read-only blast-radius analysis — use it when you want to understand scope before deciding whether to proceed.lsp-refactorsequences ALL four workflows (lsp-impact → lsp-safe-edit → lsp-verify → lsp-test-correlation) in order. Use it when you know your target and intent up front and want the complete workflow without switching skills.
Input
- target: symbol name in dot notation (e.g.
"codec.Encode","Buffer.Reset") OR file path (e.g."internal/lsp/client.go") - intent: description of the change to make (e.g. "rename to ParseConfigV2",
"add a second parameter
timeout time.Duration") - workspace_root: absolute path to the workspace root
Phase 1 — Blast-Radius Analysis (inlined from lsp-impact)
This phase is mandatory. Do not skip it, even for "small" refactors.
Call mcp__lsp__blast_radius with changed_files set to the file
containing the target symbol. If the user provided a file path directly, use it.
If the user provided a symbol name, resolve the file first (e.g. via
mcp__lsp__go_to_symbol).
mcp__lsp__blast_radius({
"changed_files": ["/abs/path/to/file"],
"include_transitive": false
})
Returns:
affected_symbols— exported symbols with reference countstest_callers— test files and enclosing test function namesnon_test_callers— production call sites
Display:
- Affected symbol count
- Test callers (each with enclosing test function name)
- Non-test callers (each with file:line)
- Total reference count
High blast-radius gate: If the total reference count exceeds 20, STOP and ask the user to confirm before continuing:
High blast radius: N callers found. Proceed with refactor? [y/n]
If the user answers "n", abort. Do not proceed to Phase 2.
Phase 2 — Speculative Preview (inlined from lsp-safe-edit)
Only reached if Phase 1 blast radius is acceptable (≤ 20 callers, or user confirmed).
2a — Open file and capture baseline diagnostics
mcp__lsp__open_document({ "file_path": "/abs/path/to/file", "language_id": "go" })
mcp__lsp__get_diagnostics({ "file_path": "/abs/path/to/file" })
Store baseline diagnostics as BEFORE.
2b — Speculative simulation
For a single-file change: use preview_edit:
mcp__lsp__preview_edit({
"file_path": "/abs/path/to/file",
"start_line": <N>,
"start_column": <col>,
"end_line": <N>,
"end_column": <col>,
"new_text": "<replacement text>"
})
For a multi-file change (e.g. rename + call site updates): use simulate_chain:
mcp__lsp__simulate_chain({
"workspace_root": "/abs/path/to/workspace",
"language": "go",
"edits": [
{
"file_path": "/abs/path/to/file.go",
"start_line": <N>, "start_column": <col>,
"end_line": <N>, "end_column": <col>,
"new_text": "<replacement>"
}
// additional dependent edits ...
]
})
2c — Evaluate simulation result
Display the speculative result using the Diagnostic Diff Output Format from references/patterns.md.
Decision:
net_delta | Action |
|---|---|
| ≤ 0 | Safe. Proceed to Phase 3. |
| > 0 | Abort. Report introduced errors. Do NOT apply to disk. |
If net_delta > 0, stop and show the full list of errors the simulation
introduced. Do not proceed to Phase 3.
Phase 3 — Apply to Disk
Only reached if Phase 2 net_delta <= 0.
Apply the change using the Edit or Write tool. When the edit targets a complete
function or method body, mcp__lsp__replace_symbol_body is an alternative that
resolves the symbol by name and replaces its full range without position math:
mcp__lsp__replace_symbol_body({
"file_path": "/abs/path/to/file",
"symbol_path": "Package.Function",
"new_body": "func Function() error {\n\treturn nil\n}"
})
For edits computed by simulation, mcp__lsp__apply_edit may be used directly if
the simulation returned an edit object:
Edit(file_path: "/abs/path/to/file", old_string: "...", new_string: "...")
For multi-file changes, apply each file's edits before moving to Phase 4. If any individual apply fails, stop and report before applying remaining files.
After applying, format the changed file(s):
mcp__lsp__format_document({ "file_path": "/abs/path/to/file" })
Apply the returned TextEdit[] via mcp__lsp__apply_edit if non-empty.
Phase 4 — Build Verification (inlined from lsp-verify)
Run in this order — LSP diagnostics first, then the compiler build:
mcp__lsp__get_diagnostics({ "file_path": "/abs/path/to/file" })
mcp__lsp__run_build({ "workspace_root": "/abs/path/to/workspace" })
Decision:
| Result | Action |
|---|---|
| Diagnostics clean, build passes | Proceed to Phase 5. |
| Diagnostics show new errors | Display errors and stop. Do not proceed to Phase 5. |
| Build fails | Display build output and stop. Do not proceed to Phase 5. |
If build fails, report the full build error output and stop. Test execution is skipped until build passes.
Phase 5 — Run Affected Tests (inlined from lsp-test-correlation)
For each file changed in Phase 3, find correlated test files:
mcp__lsp__get_tests_for_file({ "file_path": "/abs/path/to/changed/file" })
Deduplicate the resulting test files if multiple source files were changed. Run only the correlated test files:
mcp__lsp__run_tests({ "workspace_root": "/abs/path/to/workspace", "test_files": [...] })
If no correlated test files are found: note "No test correlation found —
run full suite manually to confirm." Do not attempt to run ./... automatically.
Abort Conditions
The following conditions abort the workflow immediately. Each abort displays the relevant output before stopping.
- Phase 1: blast radius > 20 callers AND user does not confirm → abort
- Phase 2:
net_delta > 0(simulation introduced errors) → abort, show errors - Phase 4: build fails → abort, show build output
- Any phase: LSP tool returns an unexpected error → abort, report tool output verbatim
Output Format
After completing all phases, produce this structured report:
## lsp-refactor Complete
### Phase 1 — Blast Radius
Affected symbols: N
Test callers: M (list each with enclosing test function)
Non-test callers: K
### Phase 2 — Speculative Preview
[Diagnostic Diff Output Format from patterns.md]
net_delta: 0 → safe to apply
### Phase 3 — Applied
Files changed: [list]
### Phase 4 — Build Verification
Diagnostics: N errors (0 new)
Build: PASS
### Phase 5 — Test Results
Test files run: [list]
Result: PASS / FAIL
If the workflow aborted at a phase, report only the phases completed and the abort reason:
## lsp-refactor Aborted at Phase 2
### Phase 1 — Blast Radius
...
### Phase 2 — Speculative Preview
ABORTED: net_delta: +2 (errors introduced)
Errors:
- file.go:34 — undefined: NewType
- file.go:51 — cannot use int as string
Example
Goal: rename exported function ParseConfig → ParseConfigV2 in pkg/config
Phase 1 — Blast Radius
blast_radius(changed_files=["pkg/config/parser.go"])
→ affected_symbols: 1 (ParseConfig)
→ non_test_callers: 3 (cmd/main.go, internal/app.go, internal/loader.go)
→ test_callers: 1 (pkg/config/parser_test.go — TestParseConfig)
→ total references: 4 — within threshold, proceeding
Phase 2 — Speculative Preview
open_document(file_path="pkg/config/parser.go")
get_diagnostics → BEFORE: 0 errors
simulate_chain(edits: [parser.go rename + 3 call-site updates])
→ cumulative_delta: 0 → safe to apply
Phase 3 — Applied
Edit parser.go: func ParseConfig → func ParseConfigV2
Edit cmd/main.go, internal/app.go, internal/loader.go: update call sites
format_document(parser.go)
Phase 4 — Build Verification
get_diagnostics → 0 errors
run_build → success
Phase 5 — Test Results
get_tests_for_file(parser.go) → pkg/config/parser_test.go
run_tests(test_files=["pkg/config/parser_test.go"]) → PASS
## lsp-refactor Complete
...
GitHub Repository
Related Skills
content-collections
MetaThis skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.
polymarket
MetaThis skill enables developers to build applications with the Polymarket prediction markets platform, including API integration for trading and market data. It also provides real-time data streaming via WebSocket to monitor live trades and market activity. Use it for implementing trading strategies or creating tools that process live market updates.
creating-opencode-plugins
MetaThis skill helps developers create OpenCode plugins that hook into 25+ event types like commands, files, and LSP operations. It provides the plugin structure, event API specifications, and implementation patterns for JavaScript/TypeScript modules. Use it when you need to intercept, monitor, or extend the OpenCode AI assistant's lifecycle with custom event-driven logic.
sglang
MetaSGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.
