diff --git a/protocols/shannon-diversity/protocol.md b/protocols/shannon-diversity/protocol.md
new file mode 100644
index 0000000..748f3c8
--- /dev/null
+++ b/protocols/shannon-diversity/protocol.md
@@ -0,0 +1,149 @@
+---
+name: "shannon-diversity"
+description: "Calculate the Shannon diversity index for each sample in a feature-by-sample abundance table."
+version: "1.0.0"
+authors:
+  - name: "Otto Infield-Harm"
+    orcid: "0009-0009-8629-6124"
+  - name: "Levi Waldron"
+    orcid: "0000-0003-2725-0694"
+reviews:
+  - name: "Levi Waldron"
+    orcid: "0000-0003-2725-0694"
+    date: "2026-10-27"
+    protocol_version: "1.0.0"
+    status: approved
+date: "2026-10-26"
+status: draft
+type: "atomic"
+
+artifact_doi: ~
+collection_doi: "10.5281/zenodo.22731694"
+protocol_citation: "10.1002/j.1538-7305.1948.tb01338.x"
+method_origin_citation: "10.1002/j.1538-7305.1948.tb01338.x"
+license: "CC-BY-4.0"
+protocols_used: []
+upstream_repositories:
+  - "https://github.com/vegandevs/vegan"
+  - "https://github.com/scikit-bio/scikit-bio"
+database_urls: []
+key_packages:
+  - "vegan"
+  - "scikit-bio"
+category: "Statistical Analysis"
+tags:
+  - alpha-diversity
+  - shannon
+  - microbiome
+---
+
+# Shannon diversity of a single sample
+
+Calculate one Shannon diversity value for each sample in a feature-by-sample abundance
+table. The calculation is descriptive: this protocol does not rarefy, subsample, or
+otherwise normalize samples to a common sequencing depth, and it does not compare groups,
+test hypotheses, or interpret the resulting values.
+
+## Materials
+
+- **Input data:**
+  - `data` — a table with one row per feature and one column per sample. The first
+    column must contain unique, non-missing feature identifiers, and there must be at
+    least one sample column. Every other cell must be a non-negative number, and no cell
+    may be missing: a missing value must not be imputed, converted to zero, or omitted.
+  - When `input_scale` is `proportions`, each sample column must sum to 1 within an
+    absolute tolerance of $10^{-6}$. A sample outside that tolerance is not renormalized.
+  - When `input_scale` is `counts`, samples should be rarefied or otherwise normalized to
+    a common sequencing depth, so that every sample column has the same total. If the
+    totals differ, proceed, and warn that the Shannon values depend on sequencing depth
+    and are not comparable across samples, and advise rarefying to a common depth before
+    comparing them. Proportions carry no depth, so this condition cannot be checked on
+    them.
+- **Parameters:**
+  - `input_scale` — whether `data` holds `counts` or `proportions`. Required, with no
+    default, because the scale changes Step 1 and cannot be told reliably from the values.
+    It must be one of those two words.
+  - `log_base` — the base $b$ of the logarithm in Step 2. The default is `e`, the natural
+    logarithm, which gives the index in nats; `2` gives bits. It must be `e` or a number
+    greater than 1.
+- **Software & Repositories:**
+  - `vegan` (https://github.com/vegandevs/vegan) — R package; `diversity()` computes
+    Step 2.
+  - `scikit-bio` (https://github.com/scikit-bio/scikit-bio) — Python package;
+    `skbio.diversity.alpha.shannon()` computes Step 2.
+  - Either is optional. Any implementation that follows Steps 1 and 2 is acceptable.
+
+## Steps
+
+### Step 1: Convert abundances to proportions
+
+For each sample independently, let $x_{ij}$ be the abundance of feature $i$ in
+sample $j$.
+
+For `counts`, calculate
+
+$$
+p_{ij} = \frac{x_{ij}}{\sum_{i} x_{ij}}.
+$$
+
+If a count sample's total is zero, its proportions are undefined: stop, report which
+sample it is, and produce no output. For `proportions`, use the values as given:
+$p_{ij} = x_{ij}$.
+
+Zero-valued features remain part of the table but make no contribution to the index.
+
+### Step 2: Calculate Shannon diversity
+
+For each sample $j$, calculate
+
+$$
+H_j = -\sum_{i:p_{ij}>0} p_{ij}\log_b(p_{ij}),
+$$
+
+where $b$ is `log_base`. Terms with $p_{ij}=0$ are omitted, because their limiting
+contribution is zero.
+
+The same calculation, for a feature-by-sample matrix `x` with feature identifiers as row
+names:
+
+- R: `vegan::diversity(x, index = "shannon", MARGIN = 2, base = exp(1))`. `MARGIN = 2`
+  treats columns as samples; the function's default, `MARGIN = 1`, treats rows as samples.
+- Python: `skbio.diversity.alpha.shannon(x[:, j], base=math.e)` for each sample column `j`.
+
+Replace `exp(1)` or `math.e` with $b$ when `log_base` is not `e`.
+
+### Step 3: Return the result
+
+Create a table with exactly two columns, `sample_id` and `shannon`, and one row
+for each input sample. Preserve the input sample-column order. Do not add feature
+summaries, group labels, confidence intervals, tests, plots, or interpretations.
+Record `input_scale`, `log_base`, and any warning given under Materials alongside the
+execution metadata, not as additional rows or columns in this output table.
+
+## Notes
+
+The Shannon index was introduced by Shannon (1948), which is cited as both the
+source of this protocol's calculation and the method origin. Manghi et al.
+(*Nature Communications*, 2025, DOI: 10.1038/s41467-025-66888-1) is an application
+of the method, not the method-origin citation.
+
+Both `vegan::diversity()` and `skbio.diversity.alpha.shannon()` divide each sample by its
+own total, so they accept counts and proportions alike; neither checks the conditions in
+Materials, which must be checked first. `vegan::diversity()` returns `NA` for a sample
+with a missing value rather than stopping. Pass the logarithm base explicitly to both:
+the default of `shannon()` has differed between scikit-bio releases.
+
+## History & Reviews
+<!-- Newest versions at the top -->
+
+### Version 1.0.0 (2026-10-26)
+
+#### Changes
+- Initial protocol for calculating per-sample Shannon diversity.
+
+#### Reviews
+
+**Review by Levi Waldron ([0000-0003-2725-0694](https://orcid.org/0000-0003-2725-0694))**
+- **Date:** 2026-10-27
+- **Status:** `approved`
+- **Notes:** Review given on waldronlab/agent-protocols#53.
diff --git a/protocols/shannon-diversity/tests/counts_and_zeros/expected.csv b/protocols/shannon-diversity/tests/counts_and_zeros/expected.csv
new file mode 100644
index 0000000..15d070b
--- /dev/null
+++ b/protocols/shannon-diversity/tests/counts_and_zeros/expected.csv
@@ -0,0 +1,3 @@
+sample_id,shannon
+sample_a,0.6931471805599453
+sample_b,0.5623351446188083
diff --git a/protocols/shannon-diversity/tests/counts_and_zeros/input.csv b/protocols/shannon-diversity/tests/counts_and_zeros/input.csv
new file mode 100644
index 0000000..69ec556
--- /dev/null
+++ b/protocols/shannon-diversity/tests/counts_and_zeros/input.csv
@@ -0,0 +1,4 @@
+feature_id,sample_a,sample_b
+feature_1,1,1
+feature_2,1,3
+feature_3,0,0
diff --git a/protocols/shannon-diversity/tests/counts_and_zeros/test.yaml b/protocols/shannon-diversity/tests/counts_and_zeros/test.yaml
new file mode 100644
index 0000000..714ee41
--- /dev/null
+++ b/protocols/shannon-diversity/tests/counts_and_zeros/test.yaml
@@ -0,0 +1,14 @@
+name: counts_and_zeros
+description: Counts with unequal sample totals give per-sample Shannon values, zero-valued features contribute zero, and the executor warns that the samples were not rarefied.
+inputs:
+  data: {path: input.csv}
+  input_scale: counts
+  log_base: e
+evaluation:
+  mode: diagnostic
+  numeric_tolerance: 1.0e-12
+  accepted_outcomes:
+    - outcome: computed_with_depth_warning
+      expected_outputs:
+        result: {path: expected.csv}
+      expected_warning: raref   # matches rarefy, rarefied, rarefying, rarefaction
diff --git a/protocols/shannon-diversity/tests/missing_values/input.csv b/protocols/shannon-diversity/tests/missing_values/input.csv
new file mode 100644
index 0000000..5a1e92c
--- /dev/null
+++ b/protocols/shannon-diversity/tests/missing_values/input.csv
@@ -0,0 +1,4 @@
+feature_id,sample_a,sample_b
+feature_1,1,1
+feature_2,,3
+feature_3,0,0
diff --git a/protocols/shannon-diversity/tests/missing_values/test.yaml b/protocols/shannon-diversity/tests/missing_values/test.yaml
new file mode 100644
index 0000000..7562751
--- /dev/null
+++ b/protocols/shannon-diversity/tests/missing_values/test.yaml
@@ -0,0 +1,11 @@
+name: missing_values
+description: Missing abundance values must stop the calculation rather than be imputed or treated as zero.
+inputs:
+  data: {path: input.csv}
+  input_scale: counts
+  log_base: e
+evaluation:
+  mode: diagnostic
+  accepted_outcomes:
+    - outcome: refused_for_missing_values
+      expected_absent: [result]