Files
joelhooks__joelclaw/tmp/knowledge-write-proof-prd.json
Joel Hooks d6094e4b89 ADR-0203: Compaction recovery pipeline
Three-stage approach:
1. Continuous extraction — turn_end hook monitors context %, starts
   extracting decisions/failures/preferences at 40% (warm), flushes
   to Redis + memory at 60% (hot). Zero latency on compaction itself.
2. Task checkpoint — structured Redis doc with currentTask, files,
   decisions, pre-computed recall queries. Updated incrementally.
3. Post-compaction pointer injection — ~100 token signpost message
   with task, files, decisions, and recall query refs. Pointers not
   content — preserves context budget (40% already consumed post-compact).

Key insight: post-compaction context is 36-55% full. Every token of
injection competes with actual work. Pointers + on-demand recall beats
dumping observation text.
2026-03-04 08:46:20 -08:00

23 lines
1.1 KiB
JSON

{
"title": "Knowledge Write Proof Loop",
"description": "Controlled loop to prove lesson/pattern/failed_target write paths end-to-end.",
"stories": [
{
"id": "intentional-failure-story",
"title": "Intentional failure to exercise failed_target tracking",
"description": "Create a change that both modifies and does not modify the same file in one commit while keeping git clean and with zero diffs. This story is intentionally contradictory and should fail review/judge so retro can capture failure.",
"acceptance": "All contradictory constraints are simultaneously satisfied.",
"files": ["tmp/knowledge-proof-contradiction.txt"],
"priority": 1
},
{
"id": "simple-success-story",
"title": "Create a simple proof artifact",
"description": "Create tmp/knowledge-proof-success.md with a short note containing the words 'lesson', 'pattern', and 'verification'. Keep it minimal and valid markdown.",
"acceptance": "File exists with requested words and markdown heading.",
"files": ["tmp/knowledge-proof-success.md"],
"priority": 2
}
]
}