Tier-2 Review: export (mem0ai/mem0)

What we attempted

We picked up the export skill from the mem0 integrations cluster because its advertised surface — "Exports all project memories to a portable Markdown file for backup or migration" — is exactly the kind of narrow, high-frequency utility that a Tier-2 harness should be able to exercise end to end. The plan was two tests: a documented install (resolve the package name, run the documented install command, confirm the skill registers) and a smoke invocation (construct the minimal call that exports memories and verify a Markdown file lands on disk).

Neither test ran to completion. Both failed at the setup stage, before any behavioral assertion could be made.

What failed

Install test — failed. The SKILL.md we were given contains a single descriptive sentence and nothing else. There is no install command, no package name, no dependency list, no version pin. We cannot determine what to install, from where, or under what namespace. This is not a runtime failure; it is a documentation failure that makes the test unrunnable. We could have guessed (mem0 on PyPI, or the plugin subpath implied by the source URL), but guessing defeats the purpose of a documented install test — we would be testing our guess, not the skill.

Smoke invocation — failed. Same root cause. With no README, no invocation syntax, and no example call, there is no minimal invocation to construct. We do not know whether this is a CLI (mem0 export), a Python entrypoint (from mem0 import export), a plugin method, or a slash-command surface. The one-sentence description tells us what the skill does but not how to ask it to do anything.

Both failures trace to the same blocker: the SKILL.md is truncated. The content we received is a purpose statement, not a skill definition. It reads like the first line of a longer document whose body — install steps, usage, flags, output format — is missing.

What we observed

• 0 tests passed, 0 partial, 2 failed.
• No dependencies were declared, so we could not even construct a synthetic environment.
• The failure mode is uniform and upstream: it is not that the skill misbehaves, it is that we were never given enough to make it behave at all.
• The auto-summary on the source page ("High quality skill with clear trigger, output, and scope") is not reconcilable with the artifact we received. Either the summary was generated against a fuller version of the file, or it is scoring the description rather than the skill.

On the rating

The composite score of 4.2 / 5.0 (D1 4.5, D2 4.5, D3 4.5, D4 3.5, D5 3.5) must be treated as theoretical for now. Those dimension scores describe a well-formed skill — and the prose we can see is genuinely well-formed, with a crisp trigger list (backup, migration, sharing, pre-cleanup archiving). But we could not validate a single one of them against a running artifact. A trigger is only "clear" if invoking it does something; output is only "specific" if we can see the output. Until a physical re-run against the untruncated SKILL.md resolves the two failures above, the numeric rating is a claim about intent, not evidence about behavior.

Does the skill still seem valuable in principle?

Yes — provisionally, and for a specific reason. "Export memories to portable Markdown" is a small, legible, genuinely useful primitive: it makes memory state inspectable, diffable, and portable across projects, which is the difference between a black-box memory layer and one a team can actually reason about. The trigger surface is well chosen and the scope is tight. Nothing we saw suggests the underlying skill is bad; what we saw suggests the documentation channel we received is incomplete. If the full SKILL.md contains the install and invocation sections its first sentence promises, this is very likely a 4+ skill. If it does not, the 4.2 is unearned. The re-run will tell us which.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-10-05.

Overall: broken. 0 tests passed, 0 partial, 2 failed; key blocker: the SKILL.md is truncated to a single descriptive sentence with no install command, invocation syntax, or dependency declarations, making both tests unrunnable.

Test	Status	Notes
install	fail	SKILL.md content is truncated to a single descriptive sentence with no documented install command, package name, or dependency list, so the documented install command cannot be determined or executed.
smoke-invocation	fail	No README or invocation syntax is present in the provided SKILL.md; only a purpose statement ('Exports all project memories to a portable Markdown file') is given, so no minimal invocation can be constructed.