Search skill packs and skills together

Coding-agent projects can drift when tasks, scope, and checks are vague. This skill gives PM templates and review rules that keep Claude, Codex, and Claude Code work bounded and verifiable.

Long agent sessions can lose decisions, file history, and next steps during compression. This skill provides structured methods and probes that preserve operational context.

Agent quality is difficult to measure because outputs vary and may have several valid forms. This skill builds repeatable rubrics, tests, gates, and monitoring.

Plugin authors need a safe way to store per-project settings without committing user-specific state. This skill explains a local Markdown and YAML frontmatter pattern for Claude, Codex, and Claude Code plugins.

Agents often need spreadsheet logic without controlling Excel. This skill guides MCP and TypeScript workflows for WorkPaper formulas, file-backed edits, and local XLSX formula reports.

Vague coding-agent tasks drift, repeat work, and report success without evidence. Ariadne Loop turns them into bounded loop specs with verifiers, stop rules, and human gates.

Custom Codex agents need precise TOML files and clear runtime settings. This skill guides role design, writes configs, and validates required fields.

Long conversations can exceed context limits, increase cost, and reduce response quality. This skill selects compaction, masking, caching, partitioning, and budgeting techniques.

Manual pull request review can miss bugs, logic errors, and repository rule violations. This skill coordinates multiple AI reviewers and validates findings before reporting concise results.

Agent systems can become unreliable when prompts, tools, history, and retrieval compete for context. This skill explains context components and practical patterns for budgeting, progressive disclosure, and validation.

Complex agent work can fail when one context window carries every task. This skill helps Claude, Codex, and Claude Code design agent systems with clear handoffs.

AI agents can mark work complete without proof, which creates false completion risk. This skill requires independent evidence before task completion is accepted.