54 lines
1.4 KiB
Markdown
54 lines
1.4 KiB
Markdown
|
# feat(ai): Add context fact management for Claude Code
|
||
|
|
|
||
|
|
## Summary
|
||
|
|
|
||
|
|
Add `core ai context` subcommands to capture and manage context facts that persist across compaction.
|
||
|
|
|
||
|
|
## Required Commands
|
||
|
|
|
||
|
|
```bash
|
||
|
|
core ai context add <fact> [--source=X] # Save a context fact
|
||
|
|
core ai context list # List current facts
|
||
|
|
core ai context extract <output> # Extract actionables from tool output
|
||
|
|
core ai context clear # Clear old context (>3h)
|
||
|
|
```
|
||
|
|
|
||
|
|
## Current Shell Scripts Being Replaced
|
||
|
|
|
||
|
|
- `claude/scripts/capture-context.sh` - Stores context facts
|
||
|
|
- `claude/scripts/extract-actionables.sh` - Extracts actionables from core CLI output
|
||
|
|
|
||
|
|
## Context Format
|
||
|
|
|
||
|
|
```json
|
||
|
|
[
|
||
|
|
{"fact": "Use Action pattern not Service", "source": "user", "ts": 1234567890},
|
||
|
|
{"fact": "FAIL: TestFoo", "source": "core go test", "ts": 1234567891}
|
||
|
|
]
|
||
|
|
```
|
||
|
|
|
||
|
|
## Extraction Patterns
|
||
|
|
|
||
|
|
The `extract` command should parse output from:
|
||
|
|
- `core go test` / `core go qa` / `core go lint` - Extract ERROR/WARN/FAIL lines
|
||
|
|
- `core php test` / `core php stan` - Extract FAIL/Error lines
|
||
|
|
- `core build` - Extract error/cannot/undefined lines
|
||
|
|
|
||
|
|
## Storage
|
||
|
|
|
||
|
|
- `~/.claude/sessions/context.json`
|
||
|
|
- Max 20 items, auto-clear after 3 hours of inactivity
|
||
|
|
|
||
|
|
## Example Usage
|
||
|
|
|
||
|
|
```bash
|
||
|
|
# Manual fact
|
||
|
|
core ai context add "User prefers UK English" --source=user
|
||
|
|
|
||
|
|
# Extract from piped output
|
||
|
|
core go test 2>&1 | core ai context extract
|
||
|
|
|
||
|
|
# List facts
|
||
|
|
core ai context list --json
|
||
|
|
```
|