auto-dream

Consolidates agent session conventions into deduplicated long-term memory entries.

1|Updated Jul 23, 2026
One-click install
npx skills add https://github.com/sanjanb/my-agent-harness --skill auto-dream-sanjanb
Or copy as Structured Prompt for Agentβ–Ό
Please help me install this Agent Skill.
Skill: auto-dream
Source: https://github.com/sanjanb/my-agent-harness/tree/main/skills/auto-dream
Command: npx skills add https://github.com/sanjanb/my-agent-harness --skill auto-dream-sanjanb

SYSTEM DOCUMENTATION & REQUIREMENTS

πŸ’‘ This Skill includes scripts (resource) components.

What problem does it solve? Agent sessions accumulate learned conventions that duplicate, contradict, and grow unbounded over time, wasting context and confusing agents. This Skill periodically consolidates the conventions.jsonl memory file into canonical, deduplicated entries between sessions. ## Core Features & Use Cases - Deduplication and Merging: Detects exact matches, same-topic entries, and contradictions, then merges them into single canonical entries with superseded references. - Stale Entry Lifecycle: Flags conventions older than 90 days as stale, excludes them from agent reads, and auto-removes entries older than 120 days. - Consolidation Reports: Generates structured reports showing entries before/after, merges, supersessions, and memory health status. - Use Case: After several coding workflows, your conventions file has 85 entries with five different phrasings of the same naming rule. Run AutoDream to collapse them into 42 canonical entries and back up the original file. ## Quick Start Run the auto-dream consolidation on my .opencode/conventions.jsonl file and generate a report of merged, superseded, and stale entries.

Frequently Asked Questions about auto-dream

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I deduplicate agent memory conventions across sessions?β–Ό

Run the consolidation process, which parses each line of conventions.jsonl as JSON, groups entries by topic and tag, and merges duplicates into single canonical entries. Exact text matches keep the newer entry, while same-topic entries with different wording are merged with a supersedes field.

What is memory consolidation for AI coding agents?β–Ό

Memory consolidation is a background process that digests session learnings into long-term conventions, analogous to REM sleep. It prevents duplicate conventions, removes stale entries, and keeps the convention file bounded so agents do not waste context reading redundant entries.

When should I run convention consolidation?β–Ό

Run it when a workflow completes, when conventions.jsonl exceeds 80 entries, after major refactors, or on demand via a manual command. It should only run between sessions, never during active workflows.

How are stale conventions handled in agent memory?β–Ό

Entries older than 90 days that are never referenced get flagged with _stale: true and a reason, then excluded from agent reads. Entries older than 120 days are automatically removed. Entries under 90 days are never removed.

Can consolidation change the meaning of my conventions?β–Ό

No. The process only consolidates wording and never changes convention meaning, overrides human-written conventions, or modifies project code. The original file is backed up to conventions.jsonl.bak before any changes are written.