skilly. Buy ad slot
All skills
Community / AGENT SKILL

batch-delivery-loop

bryanzk/MyCodexEnv
0 installs 0 GitHub stars
0

Use when a user provides source text or a source artifact and wants project-aware batch planning, reviewed per-batch implementation contracts, and sequential batch execution until complete or blocked.

BEFORE YOU INSTALL

Understand the trade-offs.

SECURITY REVIEW

Not yet assessed

Review the original instructions and requested permissions before installing.

No security review is available for this catalog entry yet.

SKILL QUALITY

Not yet assessed

How clearly the skill guides your agent, how complete its workflow is, and how you can check the outcome.

No quality assessment is available for this catalog entry yet.

The full skill.

Original instructions from the publisher’s SKILL.md

# Batch Delivery Loop

Turn a source artifact into a reviewed batch delivery path, then execute only the approved batches. This skill orchestrates existing skills; it must not reimplement lifecycle routing, committee review, or repo-specific commands.

## Required Sub-Skills

- Use `delivery-harness-framework` first for repo state, lifecycle stage, execution lane, dirty worktree, and verification gates.
- Use repo-local `AGENTS.md`, `docs/repo-index.md`, `docs/harness-state.md`, and nearby source-of-truth docs before writing plans.
- Use `dual-committee-review-loop` for batch plan review and implementation contract review.

## Inputs

- `source_text_or_path`: bounded text or local artifact to transform.
- `objective`: user-facing outcome.
- `output_dir`: default `docs/plans`.
- `target_rating`: default `10/10`.
- `max_review_rounds`: default `5`.
- `execution_mode`: `plan_only`, `contracts_only`, or `full`; default `full`.
- `stop_conditions`: optional user-provided blockers or boundaries.

## Workflow

### Phase 0: Route And Guard

1. Read repo instructions and state in the order required by `delivery-harness-framework`.
2. Classify lifecycle stage, execution lane, dirty worktree ownership, forbidden actions, and PR-gated or approval-gated actions.
3. Define Definition of Done: created docs, review threshold, execution scope, verification command(s), usability check, and stop conditions.
4. Do not call live services, deploy, delete PR-gated artifacts, mutate external systems, or read secrets unless the active user request explicitly authorizes that action.
5. Before any dual review, sanitize or bound source content according to `dual-committee-review-loop`; stop if the artifact contains secrets, credentials, private customer data, production data, or unrelated local context.

### Phase 1: Batch Plan

1. Generate one batch execution plan under `output_dir`.
2. Split batches by dependency order, blast radius, ownership boundary, and verification gate.
3. Include objective, source artifact, assumptions, out-of-scope items, stop conditions, batch order, expected files or surfaces, and verification per batch.
4. Run `dual-committee-review-loop` on the plan with `target_rating` and `max_review_rounds`.
5. If review returns a non-rating `blocked` or `stop` verdict, surface that blocker verbatim and stop. Do not treat it as a failed score.
6. If review completes but does not reach `target_rating`, stop and report the rating gap. Do not create contracts or execute batches.
7. If `execution_mode` is `plan_only`, stop after the reviewed plan and report the next safe task.

### Phase 2: Implementation Contracts

1. Generate one contract doc per approved batch.
2. Each contract must include:
   - scope
   - allowed files or surfaces
   - forbidden files and actions
   - implementation steps
   - tests or checks
   - acceptance criteria
   - rollback or revert path
   - blocker conditions
3. Run `dual-committee-review-loop` on the contract set with `target_rating` and `max_review_rounds`.
4. If review returns a non-rating `blocked` or `stop` verdict, surface that blocker verbatim and stop. Do not treat it as a failed score.
5. If review completes but does not reach `target_rating`, stop before execution and report the rating gap.
6. If `execution_mode` is `contracts_only`, stop after the reviewed contracts and report the next safe task.

### Phase 3: Execute Batches

1. Execute batches sequentially in approved order.
2. Before each batch, use `delivery-harness-framework` to re-check repo status, execution lane, current blockers, and dirty-worktree ownership (`agent_owned` vs `user_owned`) before writing.
3. Use the current repo's test and verification entrypoints; do not hardcode project-specific commands in this skill.
4. After each batch, run the focused verification named in its contract.
5. Stop on a failed gate, unclear user decision, external dependency, dirty-state conflict, safety boundary, or changed source-of-truth.
6. After the final batch, run the final verification gate and report the next safe task.

## Final Report

Return:

- Plan docs created.
- Contract docs created.
- Batches completed or blocked.
- Changed files.
- Plan and contract review evidence, including dual-committee verdicts, round counts, and blockers.
- Verification evidence with `command`, `exit_code`, `key_output`, and `timestamp`.
- Remaining risks.
- Next safe task.

## Common Mistakes

- Creating three separate skills for planning, contracts, and execution.
- Copying `delivery-harness-framework` or `dual-committee-review-loop` logic instead of requiring those skills.
- Hardcoding ShipQ paths, commands, fixtures, or business rules.
- Proceeding from plan to contracts, or contracts to execution, without meeting `target_rating`.
- Treating green tests as sufficient when the batch contract also requires doc, UI, safety, or usability checks.

Skill folder

Files included alongside SKILL.md in the publisher’s repository.