Context Engineering
Helps coding agents load the right project context at the right time, reducing guesswork and convention drift.
The skill explicitly says not to commit .env files or secrets, to verify configuration and external documentation, and to treat instruction-like content as data; it also tells the agent to ask when requirements conflict or are incomplete. The deduction is for missing least-privilege guidance, explicit MCP/external data-flow disclosure, confirmation of external effects, dependency security review, and rollback procedures.
The material is internally coherent and provides a context hierarchy, loading sequence, conflict handling, and verification checklist, making the happy path understandable. Points are deducted because abnormal cases such as unavailable tools, oversized context, or failed rules-file setup lack controlled, diagnosable error feedback; static calibration also caps this score at 10.
The audience is clear—AI coding agents—and the skill names concrete scenarios such as new sessions, task switching, and declining output quality, with examples for multiple rule-file formats and MCP services. Points are deducted because non-fit boundaries, precise semantic triggers, input/output contracts, and Chinese-language or mainland-China network fit are not adequately specified.
The documentation has clear layering with overview, use cases, procedures, anti-patterns, and verification examples; the repository README supplies installation guidance, and MIT licensing and a maintainer team are visible. Points are deducted because the skill itself lacks versioning, changelog, explicit maintenance ownership and update path, and complete dependency/MCP availability notes.
Rules files, selective context loading, relevant-file inspection, ambiguity escalation, and verification steps can directly guide agents toward better context preparation and are usable for the core task. Points are deducted because no skill-specific output examples or outcome evidence are provided, and comparative benefit over manual practice or alternatives is unverified; static calibration caps this score at 7.
A real CI workflow provides static evidence for repository-wide skill validation, evaluation-runner tests, and installation checks. Points are deducted because the supplied test fixture is unrelated to context-engineering, and no dedicated key-path tests or independent corroboration for this skill are shown; static calibration caps this score at 5.
- This assessment is based only on the supplied source files; the skill, scripts, CI, and installation flow were not executed.
- MCP integrations, external documentation, and configuration files may introduce data-leakage or prompt-injection risks; authorization scope and sources should be checked before use.
- Repository-wide CI and evaluation evidence do not demonstrate the runtime effectiveness of context-engineering itself.
- The publisher is not verified by the FollowSkills curated enterprise registry and is treated as unknown.
What does this skill do, and when should you use it?
Context Engineering is a context-management workflow for AI coding agents. It organizes rules files, specifications, source files, error output, and conversation history by persistence and task relevance. It also provides context-packing patterns, conflict handling, incomplete-requirement handling, lightweight planning, and verification checks. It fits development teams that want agents to follow project conventions, reduce hallucinated APIs, and regain focus when tasks change.
Guides an agent to create and maintain CLAUDE.md, AGENTS.md, or equivalent rules files; load specifications, relevant source files, tests, type definitions, and specific error output; structure context with Brain Dump, Selective Include, and Hierarchical Summary patterns; distinguish trusted source code from configuration, external documentation, and user content that needs verification; surface conflicts and incomplete requirements with explicit options; emit a lightweight plan for multi-step work; suggest MCP integrations such as Context7, Chrome DevTools, PostgreSQL, Filesystem, and GitHub; and finish with checks for rules coverage, real project references, and context refreshes.
- A developer starting a coding session who needs the agent to understand the stack, commands, conventions, and boundaries.
- A team whose agent is inventing APIs, ignoring project patterns, or producing lower-quality output over time.
- An engineer switching between areas of a large codebase who needs to load only the relevant specification and source files.
- A team setting up a new project for AI-assisted development and needing persistent rules and a project map.
- A developer facing a conflict between the specification and existing code, or an undefined behavior that should not be guessed.
- A maintainer who wants a repeatable approach for feeding specific test failures or build errors back to the agent.
What are this skill's strengths and limitations?
- Defines a clear hierarchy and loading order for rules, specifications, source files, errors, and conversation history.
- Provides reusable context-packing templates for keeping inputs focused and avoiding context flooding.
- Explicitly requires ambiguity and incomplete requirements to be surfaced instead of silently guessed.
- Includes trust-level guidance, anti-patterns, verification checks, and suggested MCP integrations.
- The skill is primarily instructional text; no script is shown that automatically creates rules files or loads context.
- No test suite, runtime metrics, or effectiveness evaluation is provided for this individual skill.
- MCP services are optional enrichment and depend on the client and environment configuration.
- Rules filenames and client integration methods vary by tool and may require platform-specific setup.
How do you install this skill?
This skill is distributed with the repository's 24-skill collection. Install the collection with the Skills CLI: npx skills add addyosmani/agent-skills. To browse first, use: npx skills add addyosmani/agent-skills --list. The README also provides a native Codex plugin command: codex plugin marketplace add addyosmani/agent-skills.
How do you use this skill?
After installing the skill directory in an Agent Skills-compatible client, trigger context-engineering when starting a session, switching tasks, or when agent quality declines. Example: "Use context engineering to prepare the project context before we modify the authentication module." Ask the agent to read the files it will modify, related tests, and a similar existing pattern, and to pause when it finds a conflict or missing requirement.
How does this skill compare with similar options?
The README names Superpowers and Matt Pocock's skills as comparison points for the broader Agent Skills collection, but it gives no comparison specific to the context-engineering skill.