Can this "borg" other skill/prompt repos?

[TomLucidor]

Original idea that ingest prompts and output more (if possible this can be turned into a feature request) https://github.com/aplaceforallmystuff/claude-agent-borg
Cross-reference memodb-io/Acontext#581 (comment)

[IlyaKhalafi]

The Python ZIP import does not provide the pack consolidation you described. At commit a7367d0, upload_skill_memory.py reads one encountered SKILL.md, without guaranteeing which one when an archive contains several. It parses fixed sections and does not parse pack hooks.

There is also a concrete truncation problem in that snapshot. _TEXT_MAX_LEN is 20 and is applied to procedures, script contents and reference contents. I ran the unchanged parser on synthetic ZIPs and reproduced that cap. A two-skill archive produced one parsed skill. This was a parser probe, not an end-to-end MemOS run.

Splitting a pack into one ZIP per skill would address that parser's single-skill input, but would not fix truncation or implement consolidation. MemOS also advertises smart deduplication and skill evolution. I have not evaluated the separate local-plugin skill_generate/skill_evolve paths, so this is not a claim that MemOS has no related features.

For a pack-consolidation integration, first preserve files, relative links, dependencies and source revisions. Hashes can identify byte-identical duplicates; semantic similarity should nominate candidates rather than automatically merge conflicting instructions. A useful first fixture would contain two overlapping skills, a shared reference and a hook, and require their behavior to survive import and any later merge.