A Codex skill for keeping long AI work on track with compact summaries, structured memory, and timely clarification.
Long-running AI sessions often fail for a simple reason: the important context gets buried.
ContextLine keeps the throughline of a long AI session visible:
- Summarize the live context before it becomes too large or noisy.
- Store durable facts in structured memory instead of prose blobs.
- Ask narrow clarification questions before making risky assumptions.
ContextLine is a local skill for AI coding agents such as Codex.
It does not add a database, background service, telemetry, or external API call. It is a compact workflow skill that helps an agent preserve intent, decisions, constraints, artifacts, open questions, and next actions across long conversations.
This skill is useful for:
- Product managers coordinating long AI coding sessions
- Solo builders working across many chats
- Developers who want cleaner handoffs between agents
- Teams that need lightweight project memory without adding new infrastructure
- Anyone who wants an AI agent to ask before guessing when key context is missing
| Capability | Purpose | Typical Output |
|---|---|---|
| Context Summarization | Reduce token load and preserve the current task state | Goal, current state, decisions, constraints, blockers, next action |
| Structured Memory | Make important facts easy to recall and update | Stable sections for decisions, constraints, artifacts, checks, open questions |
| Proactive Clarification | Prevent hallucinated requirements and unsafe assumptions | One to three concrete questions with a safe default assumption when possible |
User prompt:
Use ContextLine to prepare a handoff for this TutorFlow prototype.
The AI changed the lesson planner and says it is ready, but I am not sure what was verified.
Expected style:
**Context**
- Goal: Prepare a handoff for the TutorFlow lesson planner prototype.
- Current state: The AI claims the planner was changed.
- Decisions: Treat "ready" as unverified until there is run output or a screenshot.
- Constraints: Keep the next handoff short and actionable.
- Open questions: What command was run? Was the planner tested in the browser?
- Next action: Ask for verification evidence before adding new features.Paste this into your AI coding agent:
Act as ContextLine.
Summarize the current task, extract structured memory, and ask only the clarification questions needed before continuing.
Conversation or project notes:
[paste content here]
Clone or download this repository, then copy the contextline/ folder into the local skills directory used by your AI agent.
For Codex-style local skills, the target is commonly:
~/.codex/skills/contextline/
After copying, start a new agent session and ask it to use $contextline.
This public demo uses only fictional project content:
Please install the ContextLine skill from this repository:
https://github.com/happiness9527/contextline-skill
Keep the skill folder name as contextline.
After installing, show me where SKILL.md was placed.
Run from the repository root:
python3 contextline/scripts/doctor.py
python3 -m py_compile contextline/scripts/*.py
git diff --checkIf you have Codex's skill creator validation script available, you can also run:
python3 /path/to/quick_validate.py contextlineAll examples use fictional projects and fictional project details.
ContextLine is not a replacement for tests, code review, user research, or product judgment.
It helps the agent manage context and uncertainty. It should still reopen source files, rerun relevant checks, and ask for confirmation when a decision depends on missing information.
MIT. See LICENSE.