memory-bridge
Browse and compare wiki knowledge by which AI tool originally produced it. Use this skill when the user says "/memory-bridge", "browse codex memory", "what did codex know about X", "show me claude knowledge", "cross-tool memory", "what does hermes know that claude doesn't", "show me knowledge from <tool>", "compare my AI tool memories", or wants to explore knowledge gaps between tools. Works from any project. Diff mode ("what's different", "unique to codex", "gaps between tools") is the killer f
Security Assessment
Detected risks:
About memory-bridge
A cross-tool knowledge browser for an Obsidian wiki, letting a user browse and compare wiki pages by which AI tool originally produced them. It is triggered by invocations like /memory-bridge, "browse codex memory", "what did codex know about X", "show me claude knowledge", or "what does hermes know that claude doesn't". The wiki tracks source provenance in a .manifest.json file and in each page's sources: frontmatter, and this skill surfaces that metadata as a navigable view. Its standout capability is diff mode, which reveals blind spots and knowledge gaps between tools that the user may not know exist.
Before running, it resolves configuration to find OBSIDIAN_VAULT_PATH, then reads .manifest.json as the source of truth for which tool produced which pages and index.md for page titles and descriptions. It parses each source entry's source_type into a canonical tool name, mapping variants like claude_conversation, codex_rollout, hermes_memory, openclaw_daily_note, copilot_session, pi_session, document (ingest), and anything else (manual), and builds a map from each tool to the set of pages it created or updated, allowing a page to belong to multiple tools.
It supports five modes: browse (list all pages from a tool), search (pages from a tool mentioning a topic), diff (pages unique to each tool, overlap, and notable gaps, optionally between two named tools), and map (a full page-by-tool origin matrix, capped and sorted by cross-tool richness). Diff computes only_in_a, only_in_b, and shared sets and explicitly calls out the most surprising asymmetry. To stay efficient it reads only frontmatter unless the user asks for more. After producing output it optionally spawns an impl-validator subagent to sanity-check parsing and diff symmetry, and appends a line to the vault's log.md. Use it to explore what each AI tool has and hasn't contributed to your knowledge base.
FAQ
What does the skill read to determine which tool produced a page?
It reads the vault's .manifest.json as the source of truth, plus each page's sources: frontmatter, and uses index.md for titles and one-line descriptions.
What modes are available?
Browse lists all pages from a tool, search finds pages from a tool mentioning a topic, diff shows pages unique to each tool plus overlap and gaps, and map builds a full page-by-tool origin matrix.
Which tools does it recognize?
claude, codex, hermes, openclaw, copilot, pi, manual (hand-written), and ingest (wiki-ingest documents), derived by mapping each source_type value in the manifest.
What makes diff mode the killer feature?
It surfaces blind spots between tools by computing pages only in tool A, only in tool B, and shared, and it explicitly highlights the most surprising asymmetry, revealing knowledge gaps the user may not know exist.
What happens if .manifest.json is missing or empty?
The skill says so clearly and suggests running /wiki-history-ingest first, since without the manifest it has no provenance data to build the source map.
Install memory-bridge
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
ar9av/obsidian-wiki