Knowledge Persistence
Heddle learns durable knowledge while it works and recalls it in future sessions.
Overview
Preferred ticket formats, canonical commands, recurring conventions, architecture notes, and workflow rules are cataloged in .heddle/memory/ and recovered in future sessions without re-discovering context.
Memory is local, reviewable markdown. Heddle records durable memory candidates during normal work, then a bounded maintainer path folds those candidates into discoverable category catalogs such as preferences, workflows, operations, domain knowledge, and current state.
What belongs in memory
What does not belong
Memory is not scratch space. Avoid storing secrets, temporary plans, raw chat transcripts, speculative guesses, or one-off debugging notes that will quickly become stale. Durable memory should help future sessions do better without making them trust unreviewed or private information.
Memory structure
.heddle/memory/README.md
.heddle/memory/current-state/
.heddle/memory/workflows/
.heddle/memory/preferences/
.heddle/memory/domain/
.heddle/memory/operations/
.heddle/memory/history/
.heddle/memory/_maintenance/Example learning loop
Teach Heddle a reusable preference, then start a later session and ask for related work. The goal is not just to remember a fact, but to preserve operating style that changes future answers.
Whenever I ask you to create a ticket, use these sections:
problem statement, proposed approach, considered alternatives, conclusion.
# Later:
Create a ticket for maintaining doc consistency after feature updates.Commands
heddle memory status # Show catalog and candidate status
heddle memory list [path] # List memory notes
heddle memory read <path> # Read a memory note
heddle memory search <query> # Search cataloged markdown memory
heddle memory validate [--repair] # Check catalog shape
heddle memory maintain --dry-run # Inspect pending memory candidates