No presentation-layer artifact role: templates and static output land as roles=["unknown"]

[rahlk]

Is your feature request related to a problem? Please describe.

discover_artifacts classifies files against RULES (codeanalyzer/artifacts/discovery.py:13-60) into an open, kebab-case role set — dependency-manifest, tool-config, container-image, service-topology, iac, ci, env, packaging, legal, docs, script, unknown. There is no presentation-layer role. A Jinja2 template, a Django template, and everything under static/ all land as format="text", roles=["unknown"] — indistinguishable from a file the analyzer has never heard of.

Consequences today: a graph consumer cannot ask "which templates does this application ship", the artifact inventory reports an application's whole view layer as unknown, and unknown stops being a useful signal — it means both "not classified" and "classified as a thing we have no name for".

codeanalyzer-java is adding view-template artifact roles (.jsp, .xhtml, Thymeleaf .html) and a bundler-output inventory (static/, dist/) as part of its web-application theme. The roles vocabulary is shared and open across all three analyzers, and no analyzer has coined a template role yet (python: 44 rules, none; codeanalyzer-typescript rules.ts: none). Whoever lands first fixes the spelling for everyone — a term coined twice is permanently wrong.

So this issue exists to hold the python end of that thread, not to race it.

Describe the solution you'd like

• A presentation-layer role value exists in the shared vocabulary, spelled identically in all three analyzers.
• Python's Jinja2 / Django templates and static output carry it instead of ["unknown"].
• format values for template files are added alongside, so a consumer can tell a JSP from a Jinja2 file without pattern-matching the path.

Describe alternatives you've considered

Not stated in the original issue.

Additional context

Scope boundary

In scope: adopting the role and format values java coins, and adding the RULES entries that classify python's equivalents — Jinja2 (templates/**.html, *.jinja, *.j2), Django templates, and bundler/static output.

Out of scope: parsing template internals ({{ }}, {% %}) into nodes, template↔code dataflow, and view-dispatch edges from render_template / render / TemplateResponse to the template artifact. Those are separate decisions on the same theme and each is larger than this one.

Blocked on codeanalyzer-java for the names only. Nothing else about it is blocked.

The naming input python owes java

• view-template over bare template. template already means Helm chart templates and cookiecutter scaffolding in this ecosystem, and values.yaml / Chart.yaml are already service-topology here. view-template is unambiguous and reads correctly for JSP, Thymeleaf, Jinja2, and .ejs/.hbs/.vue on the TypeScript side.
• format is a separate axis from roles. Artifact carries both (codeanalyzer/neo4j/schema.py, :Artifact props), so .jsp / .xhtml / .html are new format values and one new role — not one conflated string.
• Bundler output needs its own role, not the template one. A minified dist/app.js is not a view template; it is a shipped asset. Whatever java picks for row 6 of its table should be a sibling value, and python wants the same one for static/.

Caveats and known risks

• Template discovery in python is convention, not declaration: Flask's default is templates/, Django's is per-app templates/ plus DIRS in settings, and both are configurable. RULES is glob-based, so the rule will over- and under-match; that is acceptable for a role hint and must not be presented as authoritative.
• templates/**.html and a plain docs/*.html are the same extension. Ordering inside RULES decides which wins, and the existing table already relies on that (the generic *.ini fallback sits after tox.ini on purpose).
• roles is a list, so a file can be both — no need to force a single winner where view-template and something else genuinely both apply.
• Every non-.py file gets read_bytes() + full-text capture in discover_artifacts, so widening what counts as interesting has a memory cost on large static/ trees. See The analyzer ingests its own output when -o sits inside -i, growing analysis.json exponentially #207 — the output-dir self-ingestion bug — which is the same code path and should land first.
• Additive: schema_version and the graph SCHEMA_VERSION stay 2.0.0 under the 2026-09-07 hold. Consumers gate on the analyzer version floor.

Definition of done

• Role and format spellings confirmed against codeanalyzer-java's landed work, cited by issue or PR number in this thread.
• RULES entries for Jinja2 / Django templates and for static/bundler output.
• A fixture with a templates/ directory asserts the role and format on the emitted artifact, in both projections.
• unknown is no longer produced for those files.
• codeanalyzer-typescript notified (or an issue filed there) so the third analyzer adopts the same spelling rather than coining a third.