Agenter Coder

Delegate coding tasks to a separate autonomous agent with AST validation, security scanning, and automatic retry. Supports 4 backends (Claude Code, Codex, Op...

Security Scan

The skill's requirements and instructions line up with its stated purpose (delegating coding to a separate agent); it needs an Anthropic API key, Python, and the agenter package and will run a subprocess that can read/write inside the provided working directory — review and restrict the cwd before use.

Purpose & Capability

The skill is meant to run a separate coding agent and the packaged files and declared requirements (python3, uv, primary Anthropic credential) are consistent with that purpose. Requesting ANTHROPIC_API_KEY as primaryEnv matches the default backend. The included CLI wrapper calls the agenter SDK as expected.

Instruction Scope

The SKILL.md and CLI instruct the agent to operate on a user-provided --cwd and (by default) allow writing anywhere inside that cwd. This is coherent for a code-writing tool, but it means the sub-agent will perform file reads/writes and potentially execute commands in that directory. The skill recommends sandboxing and allows --allowed-write-paths to limit writes — users must pick a safe cwd or granular allowed-write-paths.

Install Mechanism

No opaque downloads are used. The documented installer uses 'uv pip install agenter>=0.1.2' (also referenced in requirements.txt), which installs a public Python package. This is a moderate-risk, expected mechanism for this kind of skill rather than an arbitrary remote download.

Credentials

Primary credential ANTHROPIC_API_KEY is appropriate for the default backend. The documentation clearly notes that other backends (e.g., codex/OpenAI) require their own keys (OPENAI_API_KEY) if selected. The skill does not demand unrelated secrets or multiple unrelated env vars by default.

Persistence & Privilege

The skill is not marked 'always:true' and is user-invocable; it does not request persistent system-wide privileges or modify other skills. It does have the capability to write files within the provided workspace (by design), but it does not declare or require elevated platform privileges.

Assessment

This skill appears to be what it claims: a wrapper that runs an external coding agent (agenter). Before installing/using it: 1) Provide only the API keys you intend to use (ANTHROPIC_API_KEY for default; provide OPENAI_API_KEY only if you choose the codex backend). 2) Always run it with --cwd set to a safe, isolated directory (or set --allowed-write-paths) so the sub-agent cannot modify sensitive files. 3) Use --sandbox unless you explicitly need no-sandbox backends. 4) Be aware the agent will send data to the chosen model provider — don’t include secrets in prompts or workspace files. 5) Review the agenter package provenance (PyPI/project repo) and consider pinning a vetted version before installing. If you need a higher-assurance review, provide the agenter package source or confirm the exact pip package ownership and release URLs.

Instead of writing code tool-by-tool (filling your context window with file contents), delegate to a purpose-built coding agent that runs in its own process. It has its own tools, validates its output with AST parsing, and comes back with structured results. Your context stays clean.

Why this instead of coding directly

• Your context window stays clean. The sub-agent does all the file reading, editing, and bash execution in its own process. You only see the final result.
• Automatic validation and retry. Every iteration runs AST syntax checks (and optional Bandit security scans). If code has errors, the agent retries automatically — no manual back-and-forth.
• Hard budget enforcement. Set a dollar limit, token limit, or time limit. The agent stops when it hits the cap — no surprise bills.
• Backend portability. Same interface whether you're using Claude, GPT, or open-source models. Switch with one flag.

When to use

Use this skill when the user asks to:

• Write, create, or generate code for a project
• Modify, refactor, or update existing code
• Fix bugs in a codebase
• Create entire applications or components
• Generate tests for existing code

Do NOT use for: reading files, explaining code, or answering questions. Use your own tools for those — they don't need a sub-agent.

How to run

python3 {SKILL_DIR}/scripts/agenter_cli.py \
  --prompt "<the coding task>" \
  --cwd "<workspace directory>" \
  --backend "anthropic-sdk" \
  --max-cost-usd 2.0 \
  --max-iterations 5 \
  --sandbox

Parameters

Flag	Required	Default	Description
--prompt	Yes	—	The coding task. Be specific about what to build.
--cwd	Yes	—	Working directory. Use the current workspace or a subdirectory.
--backend	No	anthropic-sdk	Runtime: anthropic-sdk, claude-code, codex, or openhands.
--model	No	auto	Model override (e.g., claude-sonnet-4-20250514, gpt-5.4).
--max-cost-usd	No	unlimited	Maximum spend in USD.
--max-tokens	No	unlimited	Maximum total tokens (input + output).
--max-time-seconds	No	unlimited	Maximum wall clock time.
--max-iterations	No	5	Max validation/retry iterations.
--allowed-write-paths	No	all in cwd	Glob patterns for allowed writes (e.g., "*.py" "*.ts").
--sandbox / --no-sandbox	No	--sandbox	Sandboxed execution (recommended).
--stream	No	off	Emit NDJSON progress events for real-time updates.

Cost awareness

Set budget limits based on task complexity. Always tell the user the estimated cost.

Task type	Suggested --max-cost-usd	Suggested --max-iterations
Simple script / single file	0.50	3
Small app / multiple files	2.00	5
Complex refactoring / full project	5.00	7

Backend selection

Default to anthropic-sdk unless the user asks for a specific backend. Check {SKILL_DIR}/references/backends.md if the user asks about backend differences.

• anthropic-sdk — Default. Claude Sonnet/Opus. Works with ANTHROPIC_API_KEY or AWS Bedrock.
• claude-code — Claude Code CLI runtime. Native OS-level sandbox, battle-tested file tools.
• codex — OpenAI's gpt-5.4/gpt-5.4-mini. Requires OPENAI_API_KEY.
• openhands — Any model via litellm (including local). Must use --no-sandbox.

Reading the output

The script outputs JSON to stdout:

{
  "status": "completed",
  "summary": "Created main.py with FastAPI app and test_main.py",
  "files_modified": ["main.py", "test_main.py"],
  "files": {"main.py": "...", "test_main.py": "..."},
  "iterations": 2,
  "total_tokens": 15000,
  "total_cost_usd": 0.045,
  "total_duration_seconds": 12.3
}

Status values

Status	Meaning	What to do
completed	Task succeeded, files written to disk.	Report summary and files to user.
completed_with_limit_exceeded	Task succeeded but used more resources than configured.	Report success + warn about cost.
budget_exceeded	Stopped because budget ran out before completion.	Tell user, ask if they want to retry with higher budget.
refused	The model refused the request (safety/policy).	Report refusal reason to user.
failed	Unrecoverable error.	Report error, suggest checking logs.

After running

1. Check the status field.
2. If completed: the files are already written to disk in --cwd. Use read to inspect them if the user wants to review.
3. Report the summary, cost, and files modified to the user.
4. If failed or budget_exceeded: report the issue and ask how to proceed.