Tier-2 Review: ascii-video (NousResearch Hermes Agent Skills — Creative)

What we attempted

I pulled the ascii-video skill from the Hermes Agent Skills creative collection and ran it through the standard Tier-2 harness: (1) an install step to confirm the skill could be brought into a working environment, and (2) a smoke invocation to confirm it could actually be called and produce output. The intent was to move the composite score (4.2/5.0) from a documentation-only assessment into a verified one.

What failed

Both harness steps were skipped, not passed. Zero tests passed, zero partial, zero failed — because there was nothing executable to run.

The root cause is a single, specific failure mode: the supplied SKILL.md is truncated to one descriptive line. The entire content is:

ASCII video: convert video/audio to colored ASCII MP4/GIF.

That is a capability statement, not a skill definition. There is no install command, no README, no dependency declaration, no invocation syntax, no input/output contract beyond the phrase itself, and no example. The harness could not construct an install step because none is documented, and it could not build a minimal invocation because there is no signal for how the skill is called, what arguments it takes, or what environment it expects (ffmpeg? a Python runtime? a specific model endpoint?).

Observed dependencies: none declared. That is itself a finding — a video-to-ASCII conversion pipeline almost certainly has runtime requirements (at minimum a media decoder and an encoder), and their absence from the manifest is a documentation gap, not evidence that none exist.

What we observed

• The auto-summary on the page ("clear trigger, specific output, focused scope") is consistent with the description, but the description is all the harness received.
• The per-dimension scores tell the same story: D1 (trigger clarity 4.5), D2 (output specificity 4.5), and D3 (scope precision 4.5) are high — the one-liner is genuinely well-scoped. But D4 (self-containment 3.5) and D5 (reusability 3.5) are the weakest dimensions, and those are exactly the dimensions that require content the harness never saw.
• No install or smoke-invocation could be executed or verified. The blocker is not a runtime error, a missing dependency, or a broken script — it is that the artifact under test was not fully delivered to the test surface.

Rating status

The 4.2/5.0 composite should be treated as theoretical until a physical re-run resolves the truncation. The high D1–D3 scores are inferred from the description alone; they have not been confirmed against working output. Any downstream decision that leans on this rating (adoption, ranking, comparison against sibling skills) is currently leaning on an unverified number.

Does the skill still seem valuable in principle?

Yes — and this is the honest part. The stated capability is a clean, well-bounded, genuinely useful one: a single-purpose converter from video/audio to colored ASCII MP4/GIF. That kind of skill is exactly what a creative cluster wants — narrow trigger, deterministic output type, no scope creep. Nothing in the failure suggests the underlying idea is weak. The failure is entirely about delivery: the manifest arrived truncated, so the skill could not be exercised.

If the full SKILL.md (install command, invocation syntax, dependency list, at least one worked example) is restored, this is very likely a strong skill and the 4.2 rating may well hold up — possibly with D4/D5 rising once self-containment and reusability can be judged against real content. Recommendation: re-run Tier-2 against the complete manifest before treating this rating as final. Until then, mark it unverified.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-09-27.

Overall: broken. 0 tests passed, 0 partial, 0 failed; key blocker: the SKILL.md content is truncated to a single descriptive line with no install command, README, or invocation syntax, so both tests were skipped.

Test	Status	Notes
install	skipped	The provided SKILL.md content is truncated to a single descriptive line ('ASCII video: convert video/audio to colored ASCII MP4/GIF.') and contains no documented install command, so no install step could be executed or verified.
smoke-invocation	skipped	No README or invocation syntax is present in the supplied SKILL.md text; the only content is a one-line description, so a minimal invocation cannot be constructed or run.