implement-issue

A workflow for taking a GitHub issue—the project's tracked task—from its initial plan through implementation and a pull request, a proposed code change for review. It can continue after clarification questions.

In plain words
What is it for?
Use it to implement a specific GitHub issue, handle clarification pauses, keep long waits active, and prepare the resulting branch and pull request.
Why use it?
It keeps issue work from being left unfinished when the agent needs answers or must wait for checks. It also makes sure required results are collected before the run ends.

Skill for Claude CodeCodex

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 skills/scharissis/polako/implement-issue
Any agent
npx skills add scharissis/polako --skill implement-issue
Clone the repo
git clone --depth 1 https://github.com/scharissis/polako

Made for: Claude Code, Codex.

Per session 15 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 5,525 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 $0.00015 $0.05525
Opus 5 $0.00008 $0.02763
Sonnet 5 $0.00003 $0.01105
Haiku 4.5 $0.00002 $0.00553

Measured 2d ago against content hash 24dbcbe1f9aa, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

implement-issue 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 2d ago.

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.

skills/implement-issue/SKILL.md · 367 lines

How it starts

The opening of the file, as written. The whole thing — 367 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Implement GitHub issue #$issue

This run gets one turn

Ending your turn ends the process. There is no later turn to come back to and nothing that can wake you — not a Monitor, not a scheduled wake-up, not a background job you meant to poll, not a subagent whose result you never awaited. Whatever the worktree holds the moment you stop is all the run leaves behind: uncommitted edits, an unpushed branch, no PR. To the supervisor that is indistinguishable from a run that produced nothing, so a perfectly good issue gets parked. Interactively every one of those waits works, which is exactly why this needs saying.

So anything whose result you need is waited for inside this turn, however long it takes. Never end a turn intending to resume. A measurement worth taking is worth blocking on; one not worth blocking on should be dropped rather than deferred.

Waiting is not the same as going quiet, though. A supervisor kills and resumes a run that emits nothing for -stall — fifteen minutes by default — so a wait longer than that is polled from here, in repeated calls that keep the run visibly alive, rather than spent inside one call a watchdog cannot tell from a hang. Poll it slowly, though: a check every minute or two is already an order of magnitude inside that fifteen, and each one is a fresh turn that reloads the whole late-session context — the most expensive turns in a run, not a free keepalive. A check every second or two buys nothing over that, and is how one run spent an eighth of its tool calls on sleep and status polls (issue #217). Backgrounding the slow thing is fine; what is not is the turn ending while it is still outstanding.

Stopping on purpose is a different thing from stopping to wait. An unanswered question ends the run deliberately, flagged with awaiting-answer for a human to answer and a later run to fold in — that is this run's result, not a pause, and none of the above argues for guessing instead.

No prompts, ever

Unattended means no prompts: --allowedTools grants a fixed set of command prefixes. Anything outside it — cd among them, along with EnterWorktree or any other tool ToolSearch surfaces for moving the session's own working directory — raises a confirmation nobody is there to answer. That can hang the run outright; #138's shift log shows the other way it fails, faster and just as badly — the rejected confirmation left the model to end its turn on its own a few seconds later, "finished (ok)," having produced nothing. Either way the supervisor sees no progress and parks a perfectly good issue.

Read the full file on GitHub · 367 lines

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. 2d ago First seen · 367 lines · 15 tokens per session scan A 24dbcbe1f9aa

Subscribe to this mod's changes

implement-issue is a skill published in the GitHub repository scharissis/polako (1 stars, last pushed 2d ago), licensed MIT. It adds 15 tokens to every session and 5,525 once invoked, about $0.0001 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.