contox save

Save session work into categorized sub-contexts. This is how you tell Contox what you accomplished during a coding session so it can be enriched and stored in the project brain.

Usage

bash

contox save [summary...] [options]

Arguments

Argument	Description	Required
`summary`	A plain text summary of the session work. Multiple words are joined together.	No (unless `--json` and `--auto` are not used)

Options

Flag	Description	Required	Default
`--json`	Read structured JSON input from stdin	No	`false`
`--auto`	Auto-extract changes from a transcript	No	`false`
`-c, --category <cat>`	Category for the save	No	`implementation`

Categories

The following categories are available:

Category	Description
`architecture`	Tech stack, architecture decisions, design patterns, project structure
`conventions`	Coding conventions, style rules, naming patterns, team agreements
`implementation`	Features built, components created, APIs implemented, integrations added
`decisions`	Key decisions with rationale, trade-offs considered, constraints
`bugs`	Bugs found and how they were fixed, edge cases, workarounds
`todo`	Pending tasks, known issues, planned improvements, tech debt

Examples

Save with a simple summary:

bash

contox save "Built user authentication with JWT tokens and refresh flow"

Save with a specific category:

bash

contox save -c decisions "Chose PostgreSQL over MongoDB for relational data integrity"

Save structured data from stdin:

bash

echo '{"summary":"Added auth","changes":[{"category":"implementation","title":"JWT auth","content":"Implemented JWT authentication with refresh tokens"}]}' | contox save --json

Auto-extract from a transcript:

bash

contox save --auto

How enrichment works

When you save a session, the data is sent to the Contox V2 ingest pipeline, which:

Extracts factual claims from the session evidence
Verifies claims against actual code changes
Categorizes knowledge into the appropriate brain categories
Assigns confidence scores to each extracted item
Stores approved items in the project brain for future sessions

Notes

The --auto flag attempts to extract structured changes from the current session transcript without requiring a manual summary.
When using --json, the input must be valid JSON with a summary field and an optional changes array containing objects with category, title, and content fields.
The default category is implementation, which is the most common category for session saves.

memory

scan

On this page