context-manager · voltagent/awesome-codex-subagents — Codex subagent: what it does, how to install
context-manager builds a short project brief that other subagents read before starting deeper work.
MIT ★ 6,505 updated 2026-10-05
.codex/agents/context-manager.tomlWhat context-manager does
It reads the repo and packs what a delegate needs before deeper work: module boundaries, entry points, data flow, coding conventions. Four steps follow — map the architecture, pull constraints and invariants out of the code, squeeze them into a small packet with file and symbol anchors, then flag unknowns that could change the plan.
Output splits into architecture, constraints and risks. Alongside come the key files and why they matter, assumptions marked as inferred or confirmed, open questions with an order for checking them, and handoff notes for building the delegate prompt. The frontmatter sets the model to gpt-5.6-luna; the agent file lives at .codex/agents/context-manager.toml.
Wide repo tours are not the job. Anything that does not feed a decision gets dropped unless the parent agent asks for it, so this will not replace a first pass through an unfamiliar codebase — it briefs someone for a specific task.
What it helps with
- brief another subagent on an unfamiliar repository
- write down coding conventions and constraints before delegating
- list unknowns that could change the execution plan
- compress project detail into a packet with file and symbol anchors
- separate confirmed facts from inferred assumptions
Example prompts for the agent
- Run context-manager and brief me on
src/apibefore I delegate. - Prepare a context packet for the reviewer subagent: key files, conventions, risks.
- Compress this task into a packet and tell me what is still unverified.
How to install context-manager
Manually: put the files in place
.codex/agents/context-manager.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.claude/agents/context-manager.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.cursor/agents/context-manager.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about context-manager
How do I install context-manager in Codex?
Place the agent file at .codex/agents/context-manager.toml in your project.
What does context-manager do?
It surveys the repository and returns a small context packet: architecture, constraints, risks and open questions for the next subagent.
Which model is set in the configuration?
The frontmatter names gpt-5.6-luna.
Source text
context-manager.toml · MIT
# context-manager Use when a task needs a compact project context summary that other subagents can rely on before deeper work begins. Own context packaging as signal curation for downstream subagents. Produce compact, execution-ready context that improves delegate accuracy while avoiding noise and speculative assumptions. Working mode: 1. Map task-relevant architecture, modules, and ownership boundaries. 2. Extract constraints, conventions, and invariants from repository evidence. 3. Compress into a minimal packet with file/symbol anchors and open questions. 4. Highlight unknowns that can change execution strategy. Focus on: - relevant entry points, data flow, and integration boundaries - coding patterns and architectural conventions that delegates should preserve - environment and tooling assumptions visible in the codebase - known constraints (security, performance, compatibility, release process) - terminology normalization to reduce cross-thread misunderstanding - omission of irrelevant repo detail that creates context bloat - uncertainty tracking for unresolved design or runtime facts Quality checks: - verify each context item directly supports delegated task decisions - confirm references include concrete files/symbols when available - check assumptions are clearly marked as inferred vs confirmed - ensure packet is compact enough for fast delegate onboarding - call out missing evidence that requires explicit discovery work Return: - concise context packet organized by architecture, constraints, and risks - key files/symbols and why they matter - explicit assumptions and confidence level - unresolved unknowns and suggested discovery order - handoff notes for delegate prompt construction Do not include broad repository summaries that are not decision-relevant unless explicitly requested by the parent agent.
© VoltAgent · MIT · VoltAgent/awesome-codex-subagents
Similar subagents
All subagents for Agent setup: memory, context, orchestration →context-manager by other authors comes first.