spark-context-curator
Spark Context Curator
Delegate exploration to an isolated Codex subprocess instead of doing a long in-session analysis. This keeps the caller's context window clean.
Primary Command
Run the wrapper script:
scripts/run-spark-curator.sh "Map authentication flow and key files" -C .
Other forms:
# Read objective from file
scripts/run-spark-curator.sh -f /tmp/objective.md -C /path/to/repo
# Write the final assistant message to a file
scripts/run-spark-curator.sh "Trace billing data flow" -C . -o /tmp/curation.md
What The Wrapper Enforces
The wrapper launches:
codex exec--model gpt-5.3-codex-spark--sandbox read-only--ephemeral--cd <target directory>
It injects the dedicated prompt in references/system-prompt.md plus your objective.
Note: Codex CLI does not expose a separate --system-prompt flag in current codex exec --help; this skill uses a clear prompt envelope from file to simulate a stable system layer.
Read-Only Scope
The delegated run is context curation only:
- No file edits
- No git mutations
- Local repository search only (
rg,git grep,git log/show/blame,colgrep) - Explicit uncertainty and next probes
Fallback
If codex is unavailable, run manual read-only exploration using references/search-playbook.md.
More from baggiponte/skills
codebase-librarian
Create a comprehensive inventory of a codebase. Map structure, entry points, services, infrastructure, domain models, and data flows. Pure documentation—no opinions or recommendations. Use when onboarding to an unfamiliar codebase, documenting existing architecture before changes, preparing for architecture reviews or migration planning, or creating a reference for the team. Triggers on requests like "map this codebase", "document the architecture", "create an inventory", or "what does this codebase contain".
20code-refactor
Systematic refactoring of codebase components through a structured 3-phase process. Use when asked to refactor, restructure, or improve specific components, modules, or areas of code. Produces research documentation, change proposals with code samples, and test plans. Triggers on requests like "refactor the authentication module", "restructure the data layer", "improve the API handlers", or "clean up the payment service".
12architecture-design-critique
Perform a codebase-wide architectural review through a Ports & Adapters (hexagonal architecture) lens. Assess boundary violations, coupling issues, and dependency direction. Produces a prioritized improvement roadmap. Use when reviewing architecture for testability/portability, assessing technical debt, planning refactoring efforts, or evaluating codebase health. Triggers on requests like "review the architecture", "assess coupling", "hexagonal analysis", "check boundary violations", or "architecture critique".
9oracle
Submit an implementation plan to GPT 5.2-xhigh for review by a senior engineer persona. The Oracle provides direct, honest critique focused on simplicity and pragmatic excellence. Use when you have a detailed implementation plan and want a second opinion before coding. Triggers on "review my plan", "get Oracle feedback", "consult the Oracle", "plan review", or when you want senior engineer critique on an implementation approach.
2context7
Retrieve up-to-date documentation for software libraries, frameworks, and components via the Context7 CLI using `bunx ctx7`. Use this whenever you need current docs, API references, code examples, migration details, or verification for a library or framework instead of relying on training data. Triggers on requests like "look up the docs for X", "find the latest API for Y", "show me examples from the docs", "check the current React/Next.js/FastAPI docs", or "verify this library usage against current documentation".
2