agentkitco

Coding agents lack reliable guidance when repository commands and conventions are scattered across files. This skill builds concise AGENTS.md instructions from repository evidence and connects them to Claude Code.

Long sessions and agent switches can lose decisions, test results, and unfinished work. This skill creates a concise HANDOFF.md with verified state and actionable next steps.

Code changes can introduce security problems that routine reviews miss. This skill guides focused AI triage with evidence, realistic severity, and concrete fixes.

Code cleanup can introduce behavior changes that are difficult to detect. This skill guides small refactoring steps with baseline tests, dependency checks, and repeated verification.

Local changes can hide bugs, security gaps, and missing tests before human review. This skill guides an evidence-based review with prioritized findings and explicit verification results.

Speculative fixes can hide causes and leave bugs unresolved. This skill guides assistants through reproduction, hypothesis testing, focused changes, and evidence-based verification.

Passing claims can hide skipped tests, stale results, or checks that never exercised a change. This skill guides fresh verification and evidence-based completion reports.

Reviewers need clear reasons, behavior changes, and testing evidence before assessing a pull request. This skill turns diffs and commits into structured descriptions with review priorities and rollout risks.

Unfamiliar repositories make it difficult to find entry points, setup steps, and important dependencies. This skill guides a file-referenced onboarding map with one traced flow and clear open questions.

Dependency upgrades can introduce breaking changes and obscure existing failures. This skill guides focused migrations through release notes, baseline comparisons, minimal edits, and rollback reporting.

Review preparation is difficult when contributions are scattered across pull requests, commits, and personal notes. This skill groups that evidence into a Markdown draft and marks missing results instead of inventing impact.

Stale commands, broken paths, and vague rules make agent instructions unreliable. This skill combines local heuristic checks with evidence-based review and proposes shorter instructions before approved edits.