Code One Task Group

This documentation was generated with the assistance of AI. Please report any inaccuracies.

Implement every task and sub-task in a single task group from an implementation_plan.md-style plan, dispatching each task by its declared language/framework key to the matching <key>-code-one-task-group skill — falling back to a best-effort direct implementation when its key has no matching skill installed.

Purpose

iru-code-one-task-group is the generic, language-agnostic dispatcher iru-code calls once per plan group. It resolves each task’s language/framework key, partitions the group’s tasks into per-key buckets, hands each bucket with a matching <key>-code-one-task-group skill installed (iru-java-code-one-task-group, iru-dotnet-code-one-task-group, iru-database-code-one-task-group) off to that skill in one call — which itself captures a single quality baseline, implements the bucket’s tasks (in parallel when the group allows it), and validates the whole bucket once (iru-database-code-one-task-group is a deliberately simpler exception — see Database Code One Task Group) — and implements any remaining tasks (no resolved key, or a key with no matching group skill) directly itself, best effort, following the same implement/test/license/document/verify obligations a one-task skill would. It checks off each task/sub-task’s box in implementation_plan.md and notifies the user as it happens, so iru-code doesn’t need to track per-task progress itself.

Purpose group

Issue-to-PR Development Cycle

Invocation

/iru-code-one-task-group <group text>

Inputs

Argument Required Description Default

Group text

Yes

The full text of every task and sub-task in one plan group, each with its own language/framework tag if the plan sets one, the group’s Parallelizable verdict, and any relevant "Current code state" context.

None — this skill is always invoked with a specific group’s text, by iru-code.

Outputs

  • Source and test files created/modified for every task in the group, via the dispatched <key>-code-one-task-group skill (or implemented directly for tasks with no resolved/matching key).

  • implementation_plan.md with each task’s (and sub-task’s) checkbox checked ([ ] → [x]) and a progress note, updated live as each task completes rather than batched until the whole group finishes.

  • A user-facing notification per task/sub-task as it completes, in addition to the plan-file update.

  • A summary handed back to the caller (iru-code) covering the whole group: files touched, tests added/updated, coverage achieved, code-quality outcome, and any task that stopped on a blocker.

Execution flow

flowchart TD A[Start /iru-code-one-task-group] --> B["Step 1: resolve each task's\nlanguage/framework key; find installed\n*-code-one-task-group skills"] B --> C["Partition tasks into per-key buckets\n+ a remaining/unresolved bucket"] C --> D{Bucket has a matching\n<key>-code-one-task-group skill?} D -- yes --> E["Step 2: invoke it once for\nthe whole bucket via Skill()"] D -- no --> F["Step 3: implement each task directly,\nbest effort (impl, tests, license,\ndocs, tests, coverage, quality)"] E -- blocker reported --> G["Record blocker for this task"] E -- clean --> H["Step 4: check off task/sub-task boxes\n+ notify user, per task, as it happens"] F --> H G --> H H --> I{More buckets/tasks\nremaining?} I -- yes --> D I -- no --> J["Step 5: report the group's\noutcome to the caller"]

Dependencies

Invokes

  • Java Code One Task Group — Step 2, once for the whole bucket of tasks tagged (or defaulted to) java, via a direct Skill call — that skill captures one quality baseline, implements every task in the bucket (in parallel when the group allows it), and validates the bucket once.

  • .NET Code One Task Group — the same role as iru-java-code-one-task-group, for tasks tagged dotnet.

  • Database Code One Task Group — the same role for tasks tagged database, but deliberately simpler: always sequential, with no quality baseline or bucket-wide validation of its own.

Each of the following is invoked directly, best effort, only for tasks whose key has no matching <key>-code-one-task-group skill installed (Step 3) — via the iru-gate-runner agent, so verbose report output stays out of this skill’s own context:

  • Check License — adding license headers to touched files, if this repository has a convention for them.

  • Whichever doc-comment/test/coverage/quality skills this repository has installed for the task’s actual language/framework, falling back to that language’s own generic tooling directly if none are installed.

Invoked by

  • Code — Step 4, once per plan group, in order, via the iru-isolated-skill-executor agent so each group’s own file reads/edits/reasoning stay out of `iru-code’s own context.

Related agents

  • iru-gate-runner — spawned for the best-effort fallback path (Step 3) when a task’s language/framework has no matching <key>-code-one-task-group skill installed.

Source

SKILL.md on GitHub — the file this page was generated from.