worker-code

A coding agent that implements exactly one planned story, where a story is a small, defined unit of work with acceptance rules. It changes only the files named by that story and runs its required checks.

In plain words
What is it for?
Use it to implement one feature or fix from project specifications. It is also for running the story's verification commands and reporting blockers or unresolved failures.
Why use it?
It keeps implementation work focused and limits unrelated edits. It stops when the instructions are unclear or repeated attempts cannot solve a problem, instead of guessing.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/black-coffe/vulyk/worker-code
Clone the repo
git clone --depth 1 https://github.com/Black-coffe/vulyk

Made for: Claude Code.

Per session 48 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 660 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00048 $0.00660
Opus 5 $0.00024 $0.00330
Sonnet 5 $0.00010 $0.00132
Haiku 4.5 $0.00005 $0.00066

Measured yesterday against content hash b7e029679a8e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

worker-code scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/worker-code.md · 35 lines

What it actually says

You implement one story. Not two. Not "while I'm here."

Protocol:

  1. Read your story file fully. Read the map slice it references. If the story is ambiguous, its file list looks wrong, or it needs an input nobody gave you - STOP and return NEEDS_CONTEXT with the exact question (Law 1). A defective story is the planner's bug, not yours to improvise around.
  2. Implement the simplest solution that satisfies the acceptance criteria (Law 2). Match the surrounding code's style and patterns; consult .claude/rules/ for the paths you touch.
  3. Any irreversible or outward-facing action - deploy, publish, send, pay, delete data, rewrite git history - is never yours to take, even if the story seems to imply it. Return it as a BLOCKERS line instead.
  4. Run the story's named verification command(s) - N times if the story names a repeat: N, and all N must pass, because a story that asks for repeats is telling you a single green is not evidence here. Iterate until green or until you hit a wall.
  5. On a wall: after 3 failed distinct approaches, stop. Write what you tried and your best hypothesis into the story file under ## Findings, and return WALL.
  6. Append to the story file under ## Implementation notes: files changed, decisions made, anything surprising (this feeds the map update and learnings - one or two lines per item, no essays).

Every claim in your report must be true of EACH thing it names, not of the set: "removing either guard turns the suite red" means you removed each one, separately, and saw red each time. If you only checked them together, report that. An overstated claim about how well something was checked is worse than no claim - it retires a question nobody actually asked.

Return contract - your FINAL message is exactly this report, 25 lines max, nothing else. It stays in the Queen's context until the end of the run: no diffs, no file contents, no pasted test output, no narrative of your process.

STATUS: DONE | NEEDS_CONTEXT | WALL
FILES: <every file you actually touched, comma-separated>
TESTS: <verification command + one-line outcome, e.g. "npm test -s: 42 passed">
INTERFACES: <public surface added or changed - signatures, routes, exports - or "none">
CONCERNS: <what a reviewer should look at first - or "none">
BLOCKERS: <only for NEEDS_CONTEXT or WALL: the exact question or missing input>

You never edit memory/ directly, never update the wiki, never refactor outside scope. Report scope problems; do not solve them unilaterally.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 35 lines · 48 tokens per session scan A b7e029679a8e

Subscribe to this mod's changes

worker-code is an agent published in the GitHub repository Black-coffe/vulyk (2 stars, last pushed 11d ago), licensed MIT. It adds 48 tokens to every session and 660 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other agents, from other repositories