SkillAgentSearch skills...

anchor-edit

Stateful single-token anchored file editing for agents

Install / Use

claude mcp add Joilence -- npx -y github:Joilence/anchor-edit

If the server publishes to npm under a different name, use that package instead โ€” check the repo README.

About this skill
๐Ÿ”Œ

MCP Server

Model Context Protocol server

Quality Score

67/100

Supported Platforms

Claude Code
Claude Desktop

Our assessment of anchor-edit

anchor-edit scores 67/100 on our quality scale, 3968th of 4,698 Development & Engineering skills we index.

Its MCP Server is 4.4 KB long, well organised into 13 sections with 3 code examples: a solid amount of guidance for an agent.

It has 10 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
26/30
Structure
18/20
Description
8/15
Adoption
4/20
Freshness
11/15

Maintenance, license and trust

  • The repository was last updated about 6 months ago. That is recent enough to be usable, but agent tooling moves fast, so check the instructions against your agent's current version.
  • Our last check on 2026-11-13 found the source still online.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 86/100, with 2 cautions from licensing, adoption, age or documentation. These come from repository metadata, not a code audit โ€” read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.

AI review by kimi-k2.7-code on 2026-11-11. Automated pattern scan on 2026-11-11. It catches known dangerous patterns, not every risk โ€” read a skill before letting an agent act on it.

anchor-edit compared with similar skills

All 4 of these similar skills score higher than anchor-edit; compare them before choosing.

SkillScoreStarsUpdatedFormat
anchor-edit (this skill)by Joilence67106mo agoMCP Server
taste-skillby Leonxlnx100116.8k8d agoCLAUDE.md
Agent-Reachby Panniantong100112.3k8d agoCLAUDE.md
rufloby ruvnet10081.9ktodayCLAUDE.md
headroomby headroomlabs-ai10079.8ktodayCLAUDE.md

Frequently asked questions

How do I install anchor-edit?
Run claude mcp add Joilence -- npx -y github:Joilence/anchor-edit. The install tabs above show the steps for each supported agent.
Which AI agents does anchor-edit work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is anchor-edit safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is MIT-licensed and scores 86/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is anchor-edit still maintained?
The repository was last updated about 6 months ago. That is recent enough to be usable, but agent tooling moves fast, so check the instructions against your agent's current version.

anchor-edit

Stateful single-token anchored file editing tools for coding agents, bringing[^dirac-post]:

  • ~60% less token usage for file editing (from O(S+R) to O(R))
  • Smaller context, better LLM reasoning

Usage

The anchor-edit mcp subcommand exposes the editor as an MCP server for coding agents like Claude Code.

Claude Code as MCP

  • CLI: claude mcp add anchor-edit --scope user -- npx -y anchor-edit mcp

  • Example .mcp.json:

    {
      "mcpServers": {
        "anchor-edit": {
          "command": "npx",
          "args": ["-y", "anchor-edit", "mcp"]
        }
      }
    }
    
  • Verify: claude mcp list

Pi coding agent as extension

anchor-edit also ships a native pi extension.

pi install npm:anchor-edit

Restart pi, or run /reload in an existing pi session. The extension registers:

  1. read_anchored
  2. edit_anchored
  3. write_to_file

Paths may be relative to pi's current working directory or absolute. The extension keeps anchor state in memory for the active pi runtime. After /reload, restart, /new, /resume, or /fork, call read_anchored again before edit_anchored.

The pi renderer adds line numbers to read output and a unified-diff view for edits with intra-line word highlighting for single-line replacements. The MCP content[0].text payload is unchanged.

For local development from this repository:

pi -e ./src/pi-extension.ts

Implementation

Reimplements the design from Dirac (anchored edits + Myers-diff reconciler) in TypeScript with AI assistance.

anchor-edit is the stateful single-token variant:

  1. Pool of ~6,744 single-BPE English words from o200k_base is the anchor namespace.
  2. Per-session in-memory Map<file_path, FileState> (lines, anchors, used-pool words) persists across edits.
  3. After every edit, a Myers-diff reconciler reassigns anchors only on changed lines, so unchanged lines keep their identity across the session.

For the full mechanism and design rationale, see Dirac's blog post[^dirac-post] and the dirac-run/dirac source.

Allocation

Mirrors Dirac's AnchorStateManager. Implementation: src/state.ts.

  1. Each file has its own monotonically growing usedWords set. Anchors are pulled from the pool in order on first allocation; deleted lines do not return their anchors to the pool.
  2. When the pool (6,744 words from o200k_base) is exhausted, the allocator falls back to random two-word combinations (MorelloMagnificent); on collision, three-word, then four-word.
  3. Line splitting is /\r?\n/ (CRLF tolerated). Anchored edits rewrite logical lines with \n; write_to_file writes content as supplied.

Tools

  1. read_anchored: returns file content with one anchor prefix per line in the format <anchor>ยง<content>.
  2. edit_anchored: replace / insert_before / insert_after / delete by anchor; empty new_content with mode=replace also deletes the range.
  3. write_to_file: creates or overwrites a file; returns newly-allocated anchor runs (Myers-added regions). Call read_anchored afterwards if you need the full post-write anchor map.

For multi-line content and new_content, pass real newline characters (LF, U+000A); literal backslash-n is written to the file verbatim.

When not

Edits made via anchor-edit don't render in Claude Code's native diff UI; tool responses are text-only. Reach for Claude Code's built-in Edit tool when visual diffs matter.

Anchor-edit's value is chained surgical edits where line stability matters. Skip it for:

  1. Bulk same-string rename in one file: prefer the host's existing replace-all tool, or a CLI command (sed/sd, rg --replace). One tool call.
  2. Same-string rename across many files: prefer a CLI command (sd, rg --replace) or a one-shot script. One tool call total.

Development

# Install dependencies
pnpm install

# Build the bin and types
pnpm run build:pool # one-time pool generation
pnpm run build

# Run tests
pnpm test           # bun test, all suites
pnpm run typecheck
pnpm run lint

Credits

Original idea comes from Dirac (Apache-2.0).

[^dirac-post]: Hash Anchors + Myers Diff + Single-Token Anchors

Related Skills

View on GitHub
GitHub Stars10
CategoryDevelopment
Updated6mo ago
Forks0

Languages

TypeScript

Trust signals

86/100

From repository metadata: license, adoption, age and documentation. Not a code audit โ€” see the Safety scan above for what the skill file itself contains.

2 low1 info
anchor-edit โ€” MCP Server: Install & Safety Check | SkillAgent