Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Apply ethnographic methods including prolonged engagement, participant observation, thick description, and netnography to study cultures and communities. Use this skill when the user needs to design fieldwork with immersive observation, interpret cultural practices through thick description, study online communities via netnography, or when they ask 'how do I study a culture or community', 'what is participant observation', or 'how do I apply ethnography to online settings'.
.claude/skills/asgard-ai-platform-grad-ethnography/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 13% | 0% |
Ethnography is a qualitative methodology rooted in anthropology that involves prolonged immersion in a social setting to understand cultural meanings, practices, and social structures from the participants' perspective. The researcher becomes the primary instrument, producing "thick description" (Geertz) that interprets behavior within its cultural context. Netnography extends these principles to online communities.
IRON LAW: Ethnographic validity requires PROLONGED ENGAGEMENT — short
visits produce tourist-level understanding, not cultural insight. If
your "ethnography" is based on a few interviews over two weeks, it is
NOT ethnography.Key assumptions:
Identify the field site and negotiate entry. Clarify your role on the observer-participant continuum. Build trust over time. For netnography: identify the online community, lurk to understand norms, then participate.
Observe and participate in daily activities. Record detailed fieldnotes with descriptive (what happened), reflective (your interpretations), and methodological (research decisions) layers. Spend enough time to move past "frontstage" performances to "backstage" realities.
Move beyond thin description (surface behavior) to thick description (behavior + context + meaning). Interpret actions within the local web of significance. Use emic categories (participants' own terms) before imposing etic frameworks.
Identify cultural themes, patterns, and contradictions. Triangulate fieldnotes with interviews, documents, and artifacts. Write a narrative that conveys the culture's logic while maintaining reflexive awareness of the researcher's positionality.
markdown## Ethnographic Analysis: [Community/Culture] ### Field Context - Setting: [description of the field site] - Duration: [length of engagement] - Researcher role: [observer/participant-observer/full participant] - Access strategy: [how entry was negotiated] ### Cultural Themes | Theme | Emic Term | Observed Practices | Interpretation | |-------|-----------|-------------------|----------------| | [theme] | [local term] | [what people do] | [what it means] | ### Thick Description Excerpt > [A narrative vignette showing behavior in context with interpretive layers] ### Social Structure and Power - Key actors and roles: [who matters and why] - Norms and sanctions: [what is enforced and how] - Tensions and contradictions: [where the culture is contested] ### Researcher Reflexivity - Positionality: [how the researcher's identity shaped access and interpretation] - Impact on setting: [how the researcher's presence altered behavior] ### Implications 1. [Cultural insight from the ethnography] 2. [How findings connect to broader theoretical conversations]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 41,589 | 24,743 | -41% | 1 | 1 | 0% | 5,781 | 4,496 | -22% | 0 | 0 | — |
case-02 | fail→pass | 51,977 | 24,336 | -53% | 1 | 1 | 0% | 7,143 | 4,617 | -35% | 0 | 0 | — |
case-03 | fail→pass | 38,510 | 28,141 | -27% | 1 | 1 | 0% | 5,403 | 4,897 | -9% | 0 | 0 | — |
case-04 | fail→pass | 18,889 | 8,705 | -54% | 1 | 1 | 0% | 2,874 | 2,420 | -16% | 0 | 0 | — |
case-05 | fail→fail | 29,508 | 21,345 | -28% | 1 | 1 | 0% | 4,155 | 4,032 | -3% | 0 | 0 | — |
case-06 | fail→pass | 19,654 | 14,942 | -24% | 1 | 1 | 0% | 2,871 | 3,249 | +13% | 0 | 0 | — |
case-07 | fail→pass | 16,555 | 13,130 | -21% | 1 | 1 | 0% | 2,214 | 2,966 | +34% | 0 | 0 | — |
case-08 | fail→fail | 10,674 | 17,760 | +66% | 1 | 1 | 0% | 1,542 | 3,233 | +110% | 0 | 0 | — |
case-09 | fail→pass | 15,714 | 13,541 | -14% | 1 | 1 | 0% | 2,238 | 2,813 | +26% | 0 | 0 | — |
case-10 | fail→fail | 18,366 | 19,790 | +8% | 1 | 1 | 0% | 3,023 | 4,111 | +36% | 0 | 0 | — |
case-11 | fail→pass | 14,635 | 14,169 | -3% | 1 | 1 | 0% | 2,093 | 2,999 | +43% | 0 | 0 | — |
case-12 | fail→fail | 10,758 | 19,960 | +86% | 1 | 1 | 0% | 1,625 | 3,769 | +132% | 0 | 0 | — |
case-13 | fail→fail | 13,741 | 18,955 | +38% | 1 | 1 | 0% | 2,140 | 3,595 | +68% | 0 | 0 | — |
case-14 | fail→fail | 8,210 | 16,540 | +101% | 1 | 1 | 0% | 1,183 | 3,444 | +191% | 0 | 0 | — |
case-15 | fail→pass | 7,791 | 10,396 | +33% | 1 | 1 | 0% | 1,203 | 2,743 | +128% | 0 | 0 | — |
case-16 | fail→pass | 23,237 | 19,476 | -16% | 1 | 1 | 0% | 3,221 | 3,936 | +22% | 0 | 0 | — |
case-17 | pass→pass | 9,635 | 17,032 | +77% | 1 | 1 | 0% | 1,552 | 3,514 | +126% | 0 | 0 | — |
case-18 | pass→pass | 11,183 | 10,754 | -4% | 1 | 1 | 0% | 1,712 | 2,583 | +51% | 0 | 0 | — |
case-19 | fail→pass | 20,181 | 19,037 | -6% | 1 | 1 | 0% | 2,843 | 4,079 | +43% | 0 | 0 | — |
case-20 | pass→pass | 15,342 | 13,295 | -13% | 1 | 1 | 0% | 2,553 | 2,931 | +15% | 0 | 0 | — |
case-21 | fail→fail | 38,763 | 19,139 | -51% | 1 | 1 | 0% | 5,905 | 3,667 | -38% | 0 | 0 | — |
case-22 | pass→pass | 16,370 | 16,582 | +1% | 1 | 1 | 0% | 2,607 | 3,459 | +33% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +50 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.