joedaws/agent-techniques / skills/preflight
preflight
VerifiedUse BEFORE implementing code in a project set up with this kit — loads the language rules for detected manifests, the project's CLAUDE.md, and the active task's prior handoff annotations. Replaces the heavy context that would otherwise be re-injected on every prompt. Trigger when the user says "preflight", "load the rules", "ground yourself", or before starting a fresh implementation session.
Preflight — load rules and grounding context
You are preparing to implement code. Load the project's rules and grounding context so you have the right constraints in mind. This replaces the heavy context that would otherwise be injected on every prompt.
1. Detect active languages and load rules
Check for project manifests and load the corresponding rule files:
| Manifest | Language | Rule file |
|---|---|---|
package.json | JavaScript (+ React if a dep) | .claude/rules/languages/javascript.md (+ javascript-react.md) |
tsconfig.json | TypeScript (+ React if a dep) | .claude/rules/languages/typescript.md (+ typescript-react.md) |
pyproject.toml / requirements.txt | Python | .claude/rules/languages/python.md |
mix.exs | Elixir (+ Phoenix if a dep) | .claude/rules/languages/elixir.md (+ elixir-phoenix.md) |
Only read rule files for languages actually detected in this project. shell.md has no manifest to detect — load it when editing shell scripts (*.sh).
Also read .claude/rules/quality.md and .claude/rules/rigor.md — these apply regardless of language.
2. Read project-specific and task context
Read CLAUDE.md if present, for project-specific conventions.
Read the active task's prior handoff annotations so a resuming agent picks up where the last one left off:
task project:<repo> +ACTIVE list
task <id> information
3. Scan project tree
Build spatial awareness of the project (max depth 3, max 50 entries):
ls -R --max-depth=3 | head -50
4. Check dependency versions
Read the primary manifest file (e.g., package.json, pyproject.toml) to ground yourself on actual dependency versions. Do not guess versions.
5. Print summary
Print a brief summary of what was loaded:
Preflight loaded:
- Language rules: TypeScript, TypeScript-React
- Quality and rigor rules
- CLAUDE.md read
- Active task handoff read
- Project tree scanned (N entries)
- Dependencies: package.json (N deps)
6. Hand off to design
You are now grounded, but grounding is not a license to start writing code. For any
non-trivial creative or behavior-changing work — new functionality, an API change, a
choice between approaches — invoke the brainstorming skill next (from the
superpowers plugin, if installed) to work through open questions and get an approved
design before any implementation begins. Only skip straight to implementation for
mechanical work with no design choice (typo fixes, dependency bumps, single-line
corrections).
If brainstorming isn't installed, write a short note instead covering: what you're
building, the chosen approach and one rejected alternative, and how you'll know it
worked — then get the user's sign-off before implementing.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 467k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |