skills.rest
DenisSergeevitch
DenisSergeevitchCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

repo-task-proof-loop

Automate repo-local task proof loops with spec freeze, build, evidence, and verification.

720|45|Updated Mar 22, 2026
Search Tags:#automation#verification#codex#subagents#repo-local#task-proof-loop
One-click install
npx skills add https://github.com/DenisSergeevitch/repo-task-proof-loop --skill repo-task-proof-loop
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: repo-task-proof-loop
Source: https://github.com/DenisSergeevitch/repo-task-proof-loop/tree/main
Command: npx skills add https://github.com/DenisSergeevitch/repo-task-proof-loop --skill repo-task-proof-loop

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill automates the creation and management of a repository-local task proof loop, enabling auditable, end-to-end task execution inside a codebase by storing artifacts under .agent/tasks/<TASK_ID>/ and orchestrating a spec freeze -> build -> evidence -> verify -> fix loop with fresh verification sessions.

Core Features & Use Cases

  • Initializes a strict repo-local task folder under .agent/tasks/<TASK_ID>/ and seeds essential artifacts (spec.md, evidence.md, evidence.json, verdict.json, problems.md, and raw build/test artifacts)
  • Installs project-scoped Codex and Claude subagent templates into .codex/agents/ and .claude/agents/ to support the workflow
  • Updates repo-root guidance blocks (AGENTS.md and CLAUDE.md) with a managed, proof-first workflow description
  • Drives a task loop: spec freeze, build, evidence packing, fresh verification, minimal fixes, and re-verification until PASS
  • Supports bounded parallel helper work (explorer/worker) only after explicit user authorization, keeping evidence ownership with a single builder

Quick Start

Initialize a new task with a TASK_ID and optional seed text or file, then run the managed loop from spec freeze through verification.

Frequently Asked Questions about repo-task-proof-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create auditable task proofs for repository-level refactoring or feature development?▼

Auditable task proofs are created by initializing a strict repo-local folder under .agent/tasks/<TASK_ID>/ and orchestrating a spec freeze, build, evidence packing, fresh verification, and minimal fix loop. This ensures a canonical durable state with verifiable artifacts.

What is a spec freeze to fresh verification loop for coding tasks?▼

A spec freeze to fresh verification loop is a managed workflow that drives development from a finalized specification through building, evidence collection, fresh verification, and minimal fixes until PASS. It enforces validation and verifier freshness while avoiding production-code edits.

Can I use Codex and Claude Code subagents to automate repository task loops?▼

Yes, Codex and Claude Code subagents are supported by installing project-scoped templates into .codex/agents/ and .claude/agents/. The workflow updates repo-root guidance blocks in AGENTS.md and CLAUDE.md with a managed, proof-first development description.

Does the task proof loop support parallel worker subagents for large migrations?▼

Bounded parallel helper work using explorer or worker subagents is supported only after explicit user authorization. Evidence ownership remains with a single builder to maintain auditability during large migrations, features, or bug fixes.

What artifacts are generated when managing repo-local task proofs?▼

Generated artifacts include spec.md, evidence.md, evidence.json, verdict.json, problems.md, and raw build or test artifacts. These are seeded automatically within the .agent/tasks/<TASK_ID>/ directory to provide end-to-end development proof.

When should I not use an automated task proof loop for repository changes?▼

You should avoid using an automated task proof loop when direct production-code edits are required outside a managed verification cycle, or when bounded parallel helper work is needed without explicit user authorization for evidence ownership control.

Related Skills

Explore More Software Engineering Skills→
pjy4617pjy4617

vivado-setup

Install six agents and CLAUDE.md into a target Vivado project.

Community
Advanced
crewrigcrewrig
20

gitlab-ci

Review and harden GitLab CI/CD pipeline configurations in `.gitlab-ci.yml` files.

Official
Advanced
realsigridjinrealsigridjin
20

team

Coordinate multiple agents to decompose, execute, verify, and repair engineering tasks.

Community
Advanced
mxnyawimxnyawi

qa-engineer

Systematically test features, validate fixes, and log results in .standup workflows.

Community
Intermediate