Memory graph link
A drop-in self-improving memory + operations harness for AI agents (OpenClaw / Hermes / Claude Code / Codex). Layered memory, autonomous drift-fixing loop, evals, and one-command DR. Templates + scripts, no secrets.
npx -y skills add Walliiee/agent-harness --skill memory-graph-linkAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Surface candidate [[slug]] backlinks for a new or freshly-edited wiki entry. Scans every other wiki .md across all configured workspaces for places mentioning the entry's slug/title that don't link yet, and reports suggested insertions.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
3.2 KB, 701 tokens by cl100k_base, as published. Nobody here has run it
memory-graph-link — backlink candidate finder
When a new wiki entry lands, existing entries often already mention the same concept in prose — but without the [[slug]] link the gbrain graph layer relies on. This skill scans for those orphan mentions and reports them so the author (or a follow-up edit pass) can add proper links.
When to use
- Right after
wiki-writelands a new entry. Run on the new path; review suggestions; decide which existing entries to backlink-edit. - Quarterly graph audit of high-traffic entries (proactively find stale orphan references that should now link).
How to invoke
${OPENCLAW_HOME}/bin/memory-graph-link <path-to-entry.md> [--verbose] [--max-per-file N]
| Flag | Effect |
|---|---|
| (positional) | Path to the wiki entry. Required. |
--max-per-file N | Cap suggestions per matched file (default 3) — avoids noise on long entries |
--verbose / -v | Show derived search terms and skip-reasons |
What it does
- Read the target entry; extract slug (filename stem) and title (first
#line). - Compute search terms: slug + spaced-slug + title phrase + non-stopword title tokens longer than 4 chars (skipping tokens that are substrings of the slug).
- Scan every
*.mdunder{workspace,workspace-*}/wiki/for each agent in${OPENCLAW_HOME}/config/agents.map(excludingINDEX.mdand the target itself). - For each candidate file:
- Skip if it already contains
[[slug]]. - Match search terms (longest first; phrase matches preferred).
- Cap to
--max-per-fileper target.
- Skip if it already contains
- Print a grouped report.
Output
[memory-graph-link] N candidate(s) across M file(s) for [[<slug>]]
workspace/wiki/projects/foo.md
L42 [<matched-term>] <line preview…>
→ consider adding [[<slug>]] near these lines
Exit codes
0— clean run (may have found 0 or more suggestions)2— bad arguments / unreadable file
When NOT to use
- Suggestions are advice, not auto-edits. Apply manually.
- Match is lexical, not semantic. For semantic backlink candidates (paraphrases without the exact slug), use
qmd query "<title>"and review hits manually. - Don't run on
INDEX.md(it's an index — every entry "mentions" the wiki). The script skips this by default but a--target=INDEX.mdinvocation would be useless.
Files
- Script:
${OPENCLAW_HOME}/bin/memory-graph-link(Python, ~190 lines) - This spec:
${OPENCLAW_HOME}/skills/memory-graph-link/SKILL.md
Related
wiki-write— landing point for new entries; consider invoking memory-graph-link after wiki-write returns.qmd— semantic fallback when lexical match misses paraphrased mentions.gbrain-query— graph layer that benefits from the[[slug]]links this skill suggests.