digest

Solid

Review staged memory candidates from MEMORY_CANDIDATES.md. Surface each with promote/skip/supersede options. Does not auto-promote.

AI & Automation 11 stars 3 forks Updated today MIT

Install

View on GitHub

Quality Score: 83/100

Stars 20%
36
Recency 20%
100
Frontmatter 20%
70
Documentation 15%
48
Issue Health 10%
80
License 10%
100
Description 5%
100

Skill Content

# /digest Background curation tier. Reviews staged candidates and asks the principal to: - **promote** → write to suggested destination (concepts/, decisions/, etc.) - **skip** → mark `status: skipped`, leave in ledger - **supersede** → strike + replace an existing vault entry ## When to run - At `/close` (auto, surface candidates) - When `/body-check` reports `memory_candidate_backlog > 10` - Manually when the principal asks - On Hestia's 7-day cadence (Hestia invokes `/digest` as part of her sweep) ## Behavior 1. Read `memory/MEMORY_CANDIDATES.md`. Filter `status: staged`. 2. Group by `type` (decision / preference / doctrine / project / person / skill). 3. Surface each group to the principal: source phrase, suggested destination, confidence. 4. For each, principal answers in one word: promote / skip / supersede / defer. 5. Apply the action. Update `status` and `digested_on` fields. NEVER auto-promote. Only act on principal's word.

Details

Author
wrg32786
Repository
wrg32786/aigent-os
Created
1 months ago
Last Updated
today
Language
JavaScript
License
MIT

Integrates with

Similar Skills

Semantically similar based on skill content — not just same category