feat: add 15 microbial community records from Europe PMC batch discovery

.claude/skills/add-community-record/SKILL.md

@@ -0,0 +1,75 @@
+---
+name: add-community-record
+description: Discover and add evidence-backed CommunityMech communities through local
+  scouting, identifier allocation and validated record writing; do not promote scouting
+  stubs as records.
+---
+
+# Add a CommunityMech record
+
+Use [scout-communities](../scout-communities/SKILL.md) to build a bounded candidate
+frontier; inspect primary papers and reuse previously acquired research.
+[deep-research-community](../deep-research-community/SKILL.md) is a separate
+provider workflow whose spending requires existing authorization.
+Search community names, source identifiers, exact membership and prior decisions
+across `kb/communities/`, `data/isolates/`, `kb/taxa/`, research and history with
+ignored files included. A title overlap is only a lead, not identity proof.
+
+Read [manage-identifiers](../manage-identifiers/SKILL.md) and
+[the current schema](../../../src/communitymech/schema/communitymech.yaml).
+Allocate `CommunityMech:` IDs across **both** communities and isolates; check
+again immediately before creation to avoid another session's allocation.
+Scouting stubs under `research/scouting/stubs/` contain placeholder IDs and `_scout`
+metadata and are not valid `MicrobialCommunity` records. Build from the current
+class and nearest appropriate validated record, not historical skill examples.
+
+Establish exact community scope, verified membership/strains, source-backed
+interactions, environmental/growth context and claim evidence. Write with
+`communitymech.validation.write_validated.write_validated_community` using the
+correct root class; the default is `MicrobialCommunity`, while reusable taxon
+records use `CommonTaxon`. Preserve native key order and use the curation event
+helper plus append-only `just new-history` for actual additions.
+Run `just validate <record-path>`, `just validate-strict <record-path>`,
+`just validate-terms <record-path>`, `just validate-references-explained <record-path>`,
+`just validate-history`, `just audit-writers`, `just gen-html`,
+`just check-docs-current` and `just qc`. Reuse
+[review-communities](../review-communities/SKILL.md) for its native rubric.
+
+For an open discovery goal, maintain added/already-present/rejected/blocked
+candidate dispositions. Re-query the recorded frontier after curation and report
+no further admissible candidates only within that bounded source/query scope;
+never claim that no undiscovered microbial communities exist.
+## Scope and completion
+
+For one target, read the whole maintained record and its generated projection or
+components. For a corpus goal, first freeze an inventory of record IDs, paths,
+input hashes and applicable graph/protein elements; resume that ledger after an
+interruption. Include hidden and ignored inputs when establishing absence.
+Classify each target as checked, changed, evidence-limited, excluded, not
+applicable or blocked, with reasons and inspected sources. A sample, a provider
+report and a validator pass do not establish whole-corpus scientific completion.
+
+Use [curate-yaml-record](../curate-yaml-record/SKILL.md) for the local evidence,
+write, history and status contract; this adapter does not replace its managed
+method. Record actual agent identity. Preserve scientific status rules and
+unrelated edits. Use inspected primary evidence for mechanistic claims and exact
+active identifiers for instances; never infer causality from association or
+coverage from an empty query. A quotation is source text; interpretation belongs
+in the native notes/explanation field.
+
+Save final scientific assessments through
+[review-yaml-record](../review-yaml-record/SKILL.md),
+[record reviews](../../../docs/record-reviews.md), and the
+[local profile](../../../docs/record-review-profile.md), including their native
+persistence exceptions. Keep coverage/checkpoint accounting even when a clean
+review should not produce a bundle. Unavailable prerequisites remain blocked;
+a skipped target is not a completed enrichment. This skill authorizes no paid
+provider call by itself.
+
+After a completed goal, follow the user's authorized commit, push/PR,
+adversarial review, issue/fix, merge and branch-cleanup sequence under the current
+repository guidance. Carry forward existing authorization without asking again.
+Use fresh required checks and native merge-queue rules. Verify actual merge and
+landed content before deleting that feature branch or starting the next goal.
+Only genuine review findings become issues; a no-diff goal needs a documented
+no-op receipt rather than an artificial commit.

.claude/skills/add-uniprot-protein-examples/SKILL.md

@@ -0,0 +1,74 @@
+---
+name: add-uniprot-protein-examples
+description: Review CommunityMech taxon gene annotations and add verified UniProt
+  identifiers with evidence linking exact proteins to community roles or interactions.
+---
+
+# CommunityMech UniProt gene/protein annotations
+
+Read [the schema](../../../src/communitymech/schema/communitymech.yaml),
+[ground-causal-graph-taxa](../ground-causal-graph-taxa/SKILL.md) and the full
+community plus linked reusable taxon. The native protein route is
+`CommonTaxon.genes[]` (`GeneAnnotation`), not a new `protein_examples` field on
+community interactions. `gene_id` accepts a UniProtKB CURIE, with gene symbol,
+locus, genome, GO terms, supported role/interaction and claim evidence where
+appropriate under the current schema.
+
+Inventory communities/isolates and referenced `kb/taxa/` genes. Resolve the
+protein in the exact parent taxon and experimental context. Confirm any versioned
+genome matches the parent's declared genome and that the source establishes the
+role/interaction claimed. A gene annotation must not assert a role merely because
+the family is present or a neighbor carries it. If a community-specific protein
+example lacks a suitable native owner or relationship, record the representation
+gap instead of inserting unsupported fields.
+
+Write through `communitymech.validation.write_validated.write_validated_community`
+with `target_class="CommonTaxon"` for reusable taxa, and append truthful curation
+and repository history. Run `just validate-taxa`, `just validate-terms-taxa`,
+`just validate-history`, `just audit-writers` and `just qc`; run the appropriate
+community/reference and page-current gates when linked community content changes.
+## Protein identity and evidence
+
+Inspect live UniProt entry metadata or a release-pinned, provenance-verifiable
+native snapshot. Check primary accession and active/merged/deleted state,
+organism and strain, sequence/isoform, gene/locus mapping and source identity.
+Prefer an exact assayed protein over a convenient ortholog. A named gene alone
+is not a unique protein match. Keep family, whole complex, individual subunit,
+native protein and experimental construct distinct. UniProt identity metadata
+supports identity; primary role evidence supports participation in the record's
+mechanism. Retain the original accession and an explicit mapping rationale when
+resolving a changed identifier. Do not silently substitute species or isoforms.
+## Scope and completion
+
+For one target, read the whole maintained record and its generated projection or
+components. For a corpus goal, first freeze an inventory of record IDs, paths,
+input hashes and applicable graph/protein elements; resume that ledger after an
+interruption. Include hidden and ignored inputs when establishing absence.
+Classify each target as checked, changed, evidence-limited, excluded, not
+applicable or blocked, with reasons and inspected sources. A sample, a provider
+report and a validator pass do not establish whole-corpus scientific completion.
+
+Use [curate-yaml-record](../curate-yaml-record/SKILL.md) for the local evidence,
+write, history and status contract; this adapter does not replace its managed
+method. Record actual agent identity. Preserve scientific status rules and
+unrelated edits. Use inspected primary evidence for mechanistic claims and exact
+active identifiers for instances; never infer causality from association or
+coverage from an empty query. A quotation is source text; interpretation belongs
+in the native notes/explanation field.
+
+Save final scientific assessments through
+[review-yaml-record](../review-yaml-record/SKILL.md),
+[record reviews](../../../docs/record-reviews.md), and the
+[local profile](../../../docs/record-review-profile.md), including their native
+persistence exceptions. Keep coverage/checkpoint accounting even when a clean
+review should not produce a bundle. Unavailable prerequisites remain blocked;
+a skipped target is not a completed enrichment. This skill authorizes no paid
+provider call by itself.
+
+After a completed goal, follow the user's authorized commit, push/PR,
+adversarial review, issue/fix, merge and branch-cleanup sequence under the current
+repository guidance. Carry forward existing authorization without asking again.
+Use fresh required checks and native merge-queue rules. Verify actual merge and
+landed content before deleting that feature branch or starting the next goal.
+Only genuine review findings become issues; a no-diff goal needs a documented
+no-op receipt rather than an artificial commit.

.claude/skills/curate-causal-graphs/SKILL.md

@@ -0,0 +1,67 @@
+---
+name: curate-causal-graphs
+description: Review and curate CommunityMech ecological interaction downstream graphs
+  with evidence for each directed causal claim, preserving taxon and experimental
+  scope.
+---
+
+# CommunityMech causal interaction graphs
+
+Read [the schema](../../../src/communitymech/schema/communitymech.yaml),
+[ground-causal-graph-taxa](../ground-causal-graph-taxa/SKILL.md) and
+[evidence-curation](../evidence-curation/SKILL.md).
+The native graph is `MicrobialCommunity.ecological_interactions` with
+`EcologicalInteraction.downstream` (`InteractionDownstream`), not a copied
+TraitMech nodes/edges list. Inventory `kb/communities/*.yaml` and
+`data/isolates/*.yaml`; read reusable `kb/taxa/` records that support a claim.
+
+Review each interaction and downstream target, direction, mechanism, participant
+and organism/condition scope against inspected primary sources. Preserve stable
+interaction names and their `downstream[].target` references. Taxon membership, co-occurrence
+and a metabolite correlation do not prove a causal chain. If evidence supports
+only an interaction but not the proposed downstream link, retain the narrower
+supported claim and record the unresolved step. Do not fabricate native slots.
+The causal-edge mode of
+[deep-research-community](../deep-research-community/SKILL.md) may supply leads;
+it does not itself curate the graph or satisfy scientific review.
+
+Use `communitymech.curate.curation_event.record_curation_event` and
+`communitymech.validation.write_validated.write_validated_community`; append
+actual repository history. Run community validation/strict/term/reference gates,
+`just validate-history`, `just audit-writers`, `just qc`, then `just gen-html` and
+`just check-docs-current` for changed communities. Verify downstream references
+and citations explicitly even if a structural validator accepts them.
+## Scope and completion
+
+For one target, read the whole maintained record and its generated projection or
+components. For a corpus goal, first freeze an inventory of record IDs, paths,
+input hashes and applicable graph/protein elements; resume that ledger after an
+interruption. Include hidden and ignored inputs when establishing absence.
+Classify each target as checked, changed, evidence-limited, excluded, not
+applicable or blocked, with reasons and inspected sources. A sample, a provider
+report and a validator pass do not establish whole-corpus scientific completion.
+
+Use [curate-yaml-record](../curate-yaml-record/SKILL.md) for the local evidence,
+write, history and status contract; this adapter does not replace its managed
+method. Record actual agent identity. Preserve scientific status rules and
+unrelated edits. Use inspected primary evidence for mechanistic claims and exact
+active identifiers for instances; never infer causality from association or
+coverage from an empty query. A quotation is source text; interpretation belongs
+in the native notes/explanation field.
+
+Save final scientific assessments through
+[review-yaml-record](../review-yaml-record/SKILL.md),
+[record reviews](../../../docs/record-reviews.md), and the
+[local profile](../../../docs/record-review-profile.md), including their native
+persistence exceptions. Keep coverage/checkpoint accounting even when a clean
+review should not produce a bundle. Unavailable prerequisites remain blocked;
+a skipped target is not a completed enrichment. This skill authorizes no paid
+provider call by itself.
+
+After a completed goal, follow the user's authorized commit, push/PR,
+adversarial review, issue/fix, merge and branch-cleanup sequence under the current
+repository guidance. Carry forward existing authorization without asking again.
+Use fresh required checks and native merge-queue rules. Verify actual merge and
+landed content before deleting that feature branch or starting the next goal.
+Only genuine review findings become issues; a no-diff goal needs a documented
+no-op receipt rather than an artificial commit.

.claude/skills/ground-taxa-gtdb/SKILL.md

@@ -297,7 +297,7 @@ taxonomy entries and their id anchors do not line up.
   *Chlorobium* went unprotected — it survived earlier sweeps only because its
   recompute happened to fail (#376).
 - **`is_reclassified: true`** — GTDB uses a different name than NCBI at that
-  rank. Read it as exactly that and no more: 51 of 138 true values (37%) are a
+  rank. Read it as exactly that and no more: 53 of 141 true values (38%) are a
   dropped strain designation, a polyphyly suffix, a GTDB placeholder
   epithet, or a nomenclatural ending rather than any kind of
   reclassification (#441, #480)

docs/404.html

@@ -60,7 +60,7 @@ <h1>Page Not Found (404)</h1>
             <h2>Explore CommunityMech</h2>
             <ul>
                 <li><a href="index.html">Landing Page &amp; Overview</a></li>
-                <li><a href="browser.html">Interactive Community Browser (458 communities)</a></li>
+                <li><a href="browser.html">Interactive Community Browser (473 communities)</a></li>
                 <li><a href="text_map.html">Community Text Embedding Map</a></li>
                 <li><a href="community_umap.html">Taxonomic UMAP Embedding</a></li>
                 <li><a href="sequence_map.html">ESM-2 Protein Sequence Map</a></li>
@@ -71,7 +71,7 @@ <h2>Explore CommunityMech</h2>
             </ul>
         </div>
         <footer>
-            <p>CommunityMech Knowledge Base (458 communities) &bull;
+            <p>CommunityMech Knowledge Base (473 communities) &bull;
             Schema: <a href="schema.html"><code>https://w3id.org/communitymech</code></a> &bull;
             Source: <a href="https://github.com/CultureBotAI/CommunityMech">kb/communities/</a> &bull;
             License: BSD-3-Clause</p>