harness_delivery_loop
Guides agents through a file-backed Plan -> Work -> Review -> Release loop with explicit specs, Plans.md task contracts, verification evidence, independent review, and bounded long-running execution. Use for substantial SEOSONA work that must stay auditable across sessions or agents.
- License 许可
- MIT
- 最近更新
- 2026-10-09
- Stars
- ★ 2
- Forks
- 1
01 前言 Front Matter
- name
- harness_delivery_loop
- description
- Guides agents through a file-backed Plan -> Work -> Review -> Release loop with explicit specs, Plans.md task contracts, verification evidence, independent review, and bounded long-running execution. Use for substantial SEOSONA work that must stay auditable across sessions or agents.
03 SKILL.md 正文
Harness Delivery Loop
Overview
Use this skill when chat memory is not enough. The loop converts a request into durable files, bounded work slices, independent review, and evidence packages that survive context resets.
When To Use
- Multi-step implementation or ingestion work.
- Work that may span sessions.
- Work with high risk of scope drift.
- Work where review must be separate from implementation.
Do not use this for one-line fixes or simple questions.
Operating Loop
-
Investigate.
- Read the current repository state.
- Reuse existing memory and KI items.
- Keep unobserved facts marked as
unknown.
-
Plan.
- Write or update a spec when behavior, architecture, or scope is non-trivial.
- Write or update
Plans.mdor an equivalent task checklist. - Include acceptance criteria, dependencies, stop conditions, and validation commands.
-
Work.
- Execute one approved slice at a time.
- Keep edits scoped to the active task.
- Add tests or evidence proportional to risk.
-
Verify.
- Run focused tests first.
- Run broader gates when shared behavior, routing, or publication surfaces change.
- Record exact commands and outcomes.
-
Review.
- Review separately from implementation.
- Check spec alignment, plan alignment, security, regressions, UX, and test evidence.
- Major findings block completion.
-
Release or handoff.
- Package evidence.
- Commit/push only when explicitly requested or already authorized by the task.
- Log memory/KI updates for durable learning.
Long-Running Loop
For long-running work:
- Reload the active plan at each re-entry.
- Resolve the next incomplete task.
- Verify the task contract before acting.
- Recover memory/context.
- Execute a bounded slice.
- Record evidence and status.
- Continue, schedule, or stop based on the plan.
Common Rationalizations
| Rationalization | Reality |
|---|---|
| "The plan is in the conversation." | Chat context is volatile; file-backed plans are inspectable and resumable. |
| "Review can happen at the end." | Late review makes scope drift expensive. |
| "Unknown probably means absent." | Unknown means not observed; do not promote it to a claim. |
| "One big pass is faster." | Bounded slices reduce rollback cost and make verification meaningful. |
Red Flags
- Work starts before the task boundary is clear.
- Plans list tasks without acceptance criteria.
- Verification evidence is reconstructed from memory.
- Review approves changes without command output or file references.
- The agent pushes changes without explicit publish intent.
Verification
- [ ] Current scope is recorded.
- [ ] Each completed task has evidence.
- [ ] Review is separate from implementation.
- [ ] Remaining risks are explicit.
- [ ] Memory logging was run for major outcomes.
04 同仓库更多 Skills
web_crawler_katana
High-speed web crawling using Katana patterns for deep site mapping, URL discovery, and content extraction.
server_log_analyzer
Analyzes Nginx/Apache server logs to identify Googlebot crawl patterns, errors, and performance bottlenecks.
schema_markup_generator
Auto-generates JSON-LD structured data for any page type (Article, Product, FAQ, LocalBusiness, etc.).
internal_linking_optimizer
Analyzes and optimizes the internal link graph of a website for SEO authority distribution.
google_trends_analyzer
Detects trending topics and seasonal patterns using Google Trends data for content planning and keyword strategy.
brand_context_analyzer
Tracks brand mentions, sentiment, and online reputation across web sources.