Add codespell support with configuration and typo fixes

[yarikoptic]

Introduce codespell to catch typos in comments, docstrings, and prose. Config lives in pyproject.toml; enforcement via a new GitHub Actions workflow (read-only permissions) and a pre-commit hook.

I've introduced codespell to well over a hundred projects with generally positive feedback — the FP rate is low once the config is tuned for the codebase's vocabulary (see my write-up). This repo already has ~10 prior commits fixing typos manually, which is what motivated adding automation.

Changes

Infrastructure

• .github/workflows/codespell.yml — runs codespell on push/PR against v8 (permissions: contents: read)
• .pre-commit-config.yaml — codespell hook (v2.4.2) with tomli additional dependency so it can read pyproject.toml on Python 3.10
• pyproject.toml — [tool.codespell] section with skip patterns, ignore list, and per-word rationale

Skip patterns

• worked/ — sample outputs (graph.json/html snapshots) whose truncated string identifiers extracted from scanned repos are not typos in this codebase
• translations/ (basename glob, matches docs/translations/) — foreign-language README variants that codespell (English-only) misreads

Domain-specific ignore-words-list (each documented in-file)

Word	Reason
unparseable	Consistent alternate spelling used 35× across graphify/
bloc	Dart BLoC pattern (Business Logic Component), not "block"
inout	SQL PROCEDURE param mode / Verilog keyword
dependant	FastAPI class name shown in a README example
inh	Test-local variable holding "inherits" edge set
datas	Matches edge_datas() function name from graphify.build
nd	Short variable for graph node data
bu	Short variable for Bedrock usage dict
anc	Short variable for ancestor path
ans	Short variable for answer file path
oder, ist	German stopwords in serve.py's stopword list

Inline # codespell:ignore pragma

• tests/test_dedup.py and mirrored plan doc — the "hel" trigram assertion (shingle of "hello") is protected inline rather than whitelisted globally, so future genuine typos of "hel" stay flagged.

Typo fixes (all in comments/docstrings, no functional impact)

• Patters → Patterns in graphify/extractors/dart.py (comment)
• browseable → browsable in graphify/tree_html.py (module docstring)
• implementor → implementer in graphify/llm.py (docstring)
• pre-empts → preempts in graphify/extract.py (comment)

Applied via codespell -w wrapped in datalad run for reproducibility ([DATALAD RUNCMD] commit).

Testing

• codespell . — zero errors
• pre-commit run --files ... on touched files — ruff + codespell pass (skillgen-check is unrelated env issue in my sandbox: missing python symlink to python3)

Notes

• Configuration comment in .pre-commit-config.yaml was self-referential (said "in .pre-commit-config.yaml") — fixed to point at pyproject.toml.
• The pre-commit hook uses additional_dependencies: [tomli; python_version<'3.11'] because the project supports Python 3.10 which lacks tomllib.

---

🤖 Generated with Claude Code

[graphify-labs]

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

Graphify reviewed this change.

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

---

Graphify review — findings

This PR adds codespell spell-checking to the project, introducing a new GitHub Actions workflow (.github/workflows/codespell.yml), a pre-commit hook entry, and a [tool.codespell] configuration block in pyproject.toml (with skip paths, ignore-words list, and per-word rationale comments). It applies the resulting spelling corrections to comments, docstrings, and prompt text across extract.py, llm.py, tree_html.py, and dart.py, and adds # codespell:ignore inline directives in test_dedup.py and the plan doc. The changes are confined to documentation/comment text, string content, and tooling config rather than executable logic.

No blocking issues surfaced.

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 1796 functions depend on the 463 functions this change touches.

Health — grade A; 10 existing hotspot(s) in the area this change touches (pre-existing, not introduced here):

• extract() — 359 callers, 29 callees (high)
• _rebuild_code() — 68 callers, 49 callees (high)
• deduplicate_entities() — 49 callers, 21 callees (high)
• extract_files_direct() — 15 callers, 19 callees (high)
• build_merge() — 29 callers, 8 callees (high)
• extract_xaml() — 19 callers, 12 callees (high)
• dispatch_command() — 2 callers, 110 callees (high)
• _call_claude_cli() — 26 callers, 8 callees (high)
• …and 2 more

Verification — 1796 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 1656 function(s) in the blast radius were not formally verified this run