[NaNIsaac]

Tier: low.

Brings conventions.py comment checks (BACKLOG rule 27) to 0 findings for this repo. Comments and docstrings only; no code, string literal, README, config or lock file changed.

Findings (line hits, checker run on Python 3.15, which this repo needs to parse; older Pythons skip the files that fail to parse and under-count):

rule	before	after
Korean comments or docstrings	33	0
numbered step comments	22	0
banners	24 (12 blocks: 6 in Python, 6 in Rust)	0
Note: / Important:	2	0

What was done:

• Deleted, roughly four fifths: all 12 banner blocks (blank line kept where it separated blocks, the section labels removed with them), step numbers and comments that restate the next line.
• Rewritten in English, about ten: the ones that carry a reason (why targets are passed explicitly in tree, why a target is resolved first, why relative imports are not cached, why the f-string in tags.file starts at column 0, the anchor limit in semdiff, and the like). Two Note: prefixes dropped, the sentences kept.
• Two Korean docstrings: one deleted (_is_included, restated the name), one shortened (generate_toon, a placeholder).
• Also removed two English AI-style comments next to the touched lines: the empty docstring The Master Method: and # Helper for quick access.
• No docstring feeds CLI help or __doc__: none of the changed docstrings is on a command function, so help text is unchanged.

Korean strings left alone (string literals, outside the rule): test fixtures with Hangul file names or text in tests/test_agent_use.py, tests/test_remote_scan.py, tests/test_shadow.py, rust/src/output.rs, rust/src/shadow.rs, rust/src/term.rs, rust/tests/ingest.rs, rust/tests/shadow.rs. No Korean string in src/ was touched.

Proof that only comments changed:

• Python: for all 14 touched .py files, ast.dump of the module with first-statement docstrings stripped is identical before and after (against origin/main).
• Rust: in the 5 touched .rs files every added or removed line is a // comment, except one blank line per removed banner (removed so rustfmt does not see two blank lines in a row). cargo build succeeds as before.

Local results (Windows, PYTHONUTF8=1):

• uvx ruff@0.17.1 check ok; ruff format --check ok (100 files).
• pytest -q (Python 3.15): 294 passed, 1 skipped.
• cargo fmt --check ok; cargo clippy --all-targets -- -D warnings ok; cargo test all passed; parity run (CODIGEST_PARITY=1 cargo test --no-fail-fast -- --include-ignored) all passed.
• README: the marked steps are install commands that fetch from remote main and change the machine, so they were not run here; readme-test.ps1 -DryRun lists them and passes. README is not changed by this PR.
• site/scripts/check_reference.py (docs workflow) fails with the same output on origin/main and on this branch (undocumented cdg scan options in ko/commands.md); not touched here.

[NaNIsaac]

local-ci: cd9d067 | ruff ok | pytest 294 passed | cargo ok | readme dry-run ok (install step not run)

[NaNIsaac]

The failing parity test (tests/cap.rs, parity_term_renders_what_either_build_stored) is not caused by this PR: it is a timing flake, tracked in #42. The Rust build truncates the command time to whole milliseconds and the Python build rounds it, so a command under 1 ms gets no duration attribute in one build and duration=1 in the other. The same test failed on earlier runs that do not touch it (39854272241, 39855357519, and 39924534947 on main). This PR changes comments and docstrings only; no line in rust/src/runner.rs, rust/src/term.rs or the term code changed.