docs(src): one-line XML docs, no docs on internal members #184

Grandiras · 2026-09-27T16:56:00Z

Grandiras commented

2026-09-27 16:56:00 +00:00

Part of #180 (pass F: XML docs).

What changed

Applies the lean doc rule ("Keep CS1591 everywhere. Apply the lean doc rule instead") across src/**, outside the files Pass D (shared primitives) owns.

src/ATProtoNet/Lexicon/** (models and clients). Collapsed 475 multi-line <summary> blocks that were just word-wrapped prose into the single physical line the rest of the codebase already uses for one-thought summaries (matching the convention in AtProtoClient.cs, Http/*.cs, etc.). The 4 blocks with genuine multi-paragraph <para> rationale (the chat client class docs) were left alone. Also fixed 53 public property docs that were pure restatements of the declared name ("The actors.", "The name.", …): the ~31 that are the wire-facing twin of a CursorPage<T>.Items override now say "This page's X." (real information the name/type didn't carry — that it's paginated); the rest got a one-line, context-specific description (e.g. "The list's display name.", "The list the page's members belong to.").
Internal/private XML docs, everywhere in scope. Found and processed 574 /// doc blocks on members whose effective accessibility is internal or private — including public/protected members of an internal/private enclosing type, which the compiler also doesn't require docs for. ~93 were pure boilerplate ("Request body for X.", "Response from X.", with nothing else) and were deleted outright; the rest were converted to plain // comments, stripping the XML tags to plain prose (<see cref>/<paramref>/<c> unwrapped, <param>/<returns>/<exception> folded into labeled lines) so the rationale survives without pretending to be public API documentation. Bare <inheritdoc/> with no other content was dropped entirely (7 cases — Validate() overrides and similar).

This pass is intentionally conservative about what counts as internal: a scanner tracks brace-depth and container kind (interface/enum members are implicitly public and were left alone; class/struct/record members with no explicit modifier are implicitly private) so public members nested in an internal type are caught without false-flagging real public API or interface/enum members.
Duplicated rationale outside Pass D's files. Checked that Pass A's stream-contract consolidation (EventStreamLoop/StreamConsumerOptions) and Pass B's replay-store consolidation (IJtiReplayStore) are intact — they are, and nothing needed redoing. Found and fixed one more case: MessageRecordEmbedView.Record (chat.bsky) repeated RecordEmbedView.Record's (app.bsky) sentence verbatim, since both wrap the exact same EmbeddedRecordView union; it now uses <inheritdoc cref="RecordEmbedView.Record"/>, a pattern already used elsewhere in this codebase (PdsAdminClient.cs, XrpcClient.cs).

What was left alone, and why

A handful of other near-duplicate passages exist (two Aspire With*PdsAsync extension methods share a couple of sentences inside otherwise-different <remarks> paragraphs; two stream consumers' LastSeq properties are identical). These weren't extracted:
- The Aspire duplication is a few sentences embedded mid-paragraph in otherwise genuinely different text (different config keys, different admin-account behavior for Tranquil vs. the reference PDS) — pulling just the shared sentences out via <inheritdoc>'s path filter would fragment already-careful prose for a few lines saved.
- The two LastSeq properties each contain a self-referencing <see cref="ConsumeAsync"/> — using <inheritdoc> across the two sibling classes would carry a cref that the compiler pre-resolves to the source type's method, silently pointing at the wrong type in generated docs. Not worth the correctness risk for two short properties.
- Everything else the duplicate-sentence scan surfaced (the Unknown* union variant boilerplate, sibling <param> docs, DbContext constructor boilerplate) is either the established per-type pattern for CS1591 (each type needs its own summary; the interface it implements, IUnknownUnionVariant, is a Pass D file) or standard per-overload/per-context <param> text that isn't practically shareable through XML doc tooling.
The ~200 line estimate for the Lexicon summary collapse turned out low once "leftover docs that restate names" was in scope too; the actual change is larger (net ~2000 fewer doc/comment lines) because it also covers the client-method and class-level summaries, not just DTO properties, and because the internal/private sweep (item 2) touched far more of src/** than just Lexicon/.

Verification

Comment-only check (per the brief, same method as #139): a script parses git diff -U0 origin/main, and for every changed line in a .cs file requires it to be blank, //, /*, */, or inside a block comment. Result: 0 non-comment changed lines across all three commits.
loc.py origin/main HEAD: code lines unchanged (26,951 → 26,951, confirming comment-only). XML-doc lines 14,694 → 11,415 (−3,279). Plain // comment lines 871 → 2,159 (+1,288, from the internal/private conversions). Net doc+comment lines: 15,565 → 13,574 (−1,991).
Build: dotnet build -p:EnableSourceControlManagerQueries=false — 0 warnings, 0 errors (CS1591 still an error and satisfied).
Unit tests: dotnet test tests/ATProtoNet.Tests/ -p:EnableSourceControlManagerQueries=false — 5000 passed, 0 failed, 5 skipped (Aspire manifest tests, pre-existing skip condition), run in Release-equivalent config both mid-pass and at the end.

Deferred

Nothing from this pass's assigned scope was deferred. The near-duplicates noted above under "what was left alone" were evaluated and deliberately not touched, not skipped for lack of time.

🤖 Generated with Claude Code

Part of #180 (pass F: XML docs). ## What changed Applies the lean doc rule ("Keep CS1591 everywhere. Apply the lean doc rule instead") across `src/**`, outside the files Pass D (shared primitives) owns. 1. **`src/ATProtoNet/Lexicon/**` (models and clients).** Collapsed 475 multi-line `<summary>` blocks that were just word-wrapped prose into the single physical line the rest of the codebase already uses for one-thought summaries (matching the convention in `AtProtoClient.cs`, `Http/*.cs`, etc.). The 4 blocks with genuine multi-paragraph `<para>` rationale (the chat client class docs) were left alone. Also fixed 53 public property docs that were pure restatements of the declared name (`"The actors."`, `"The name."`, …): the ~31 that are the wire-facing twin of a `CursorPage<T>.Items` override now say `"This page's X."` (real information the name/type didn't carry — that it's paginated); the rest got a one-line, context-specific description (e.g. `"The list's display name."`, `"The list the page's members belong to."`). 2. **Internal/private XML docs, everywhere in scope.** Found and processed 574 `///` doc blocks on members whose effective accessibility is internal or private — including `public`/`protected` members of an `internal`/`private` enclosing type, which the compiler also doesn't require docs for. ~93 were pure boilerplate ("Request body for X.", "Response from X.", with nothing else) and were deleted outright; the rest were converted to plain `//` comments, stripping the XML tags to plain prose (`<see cref>`/`<paramref>`/`<c>` unwrapped, `<param>`/`<returns>`/`<exception>` folded into labeled lines) so the rationale survives without pretending to be public API documentation. Bare `<inheritdoc/>` with no other content was dropped entirely (7 cases — `Validate()` overrides and similar). This pass is intentionally conservative about *what* counts as internal: a scanner tracks brace-depth and container kind (interface/enum members are implicitly public and were left alone; class/struct/record members with no explicit modifier are implicitly private) so `public` members nested in an `internal` type are caught without false-flagging real public API or interface/enum members. 3. **Duplicated rationale outside Pass D's files.** Checked that Pass A's stream-contract consolidation (`EventStreamLoop`/`StreamConsumerOptions`) and Pass B's replay-store consolidation (`IJtiReplayStore`) are intact — they are, and nothing needed redoing. Found and fixed one more case: `MessageRecordEmbedView.Record` (chat.bsky) repeated `RecordEmbedView.Record`'s (app.bsky) sentence verbatim, since both wrap the exact same `EmbeddedRecordView` union; it now uses `<inheritdoc cref="RecordEmbedView.Record"/>`, a pattern already used elsewhere in this codebase (`PdsAdminClient.cs`, `XrpcClient.cs`). ## What was left alone, and why - A handful of other near-duplicate passages exist (two Aspire `With*PdsAsync` extension methods share a couple of sentences inside otherwise-different `<remarks>` paragraphs; two stream consumers' `LastSeq` properties are identical). These weren't extracted: - The Aspire duplication is a few sentences embedded mid-paragraph in otherwise genuinely different text (different config keys, different admin-account behavior for Tranquil vs. the reference PDS) — pulling just the shared sentences out via `<inheritdoc>`'s `path` filter would fragment already-careful prose for a few lines saved. - The two `LastSeq` properties each contain a *self-referencing* `<see cref="ConsumeAsync"/>` — using `<inheritdoc>` across the two sibling classes would carry a `cref` that the compiler pre-resolves to the *source* type's method, silently pointing at the wrong type in generated docs. Not worth the correctness risk for two short properties. - Everything else the duplicate-sentence scan surfaced (the `Unknown*` union variant boilerplate, sibling `<param>` docs, `DbContext` constructor boilerplate) is either the established per-type pattern for CS1591 (each type needs its own summary; the interface it implements, `IUnknownUnionVariant`, is a Pass D file) or standard per-overload/per-context `<param>` text that isn't practically shareable through XML doc tooling. - The `~200 line` estimate for the Lexicon summary collapse turned out low once "leftover docs that restate names" was in scope too; the actual change is larger (net ~2000 fewer doc/comment lines) because it also covers the client-method and class-level summaries, not just DTO properties, and because the internal/private sweep (item 2) touched far more of `src/**` than just `Lexicon/`. ## Verification - **Comment-only check** (per the brief, same method as #139): a script parses `git diff -U0 origin/main`, and for every changed line in a `.cs` file requires it to be blank, `//`, `/*`, `*/`, or inside a block comment. Result: **0 non-comment changed lines** across all three commits. - **`loc.py origin/main HEAD`**: code lines unchanged (26,951 → 26,951, confirming comment-only). XML-doc lines 14,694 → 11,415 (−3,279). Plain `//` comment lines 871 → 2,159 (+1,288, from the internal/private conversions). Net doc+comment lines: 15,565 → 13,574 (−1,991). - **Build**: `dotnet build -p:EnableSourceControlManagerQueries=false` — 0 warnings, 0 errors (CS1591 still an error and satisfied). - **Unit tests**: `dotnet test tests/ATProtoNet.Tests/ -p:EnableSourceControlManagerQueries=false` — 5000 passed, 0 failed, 5 skipped (Aspire manifest tests, pre-existing skip condition), run in Release-equivalent config both mid-pass and at the end. ## Deferred Nothing from this pass's assigned scope was deferred. The near-duplicates noted above under "what was left alone" were evaluated and deliberately not touched, not skipped for lack of time. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

Grandiras added 3 commits

2026-09-27 16:56:00 +00:00

docs: collapse wrapped Lexicon summaries, drop internal/private XML docs (#180 ) 077ed4a59c

Pass F (XML-doc pass) of the pre-release debloat cycle:
- Lexicon models and clients: collapse ~475 multi-line <summary> blocks that
  were just wrapped prose into the single-physical-line form used elsewhere
  in the codebase.
- Every internal/private member across src/** (outside Pass D's files):
  drop XML docs that only restate the declaration, or convert the ones that
  carry real rationale into plain // comments. This also covers members
  declared public/protected whose effective accessibility is internal
  because an enclosing type is internal or private.

Comment/whitespace-only change verified with a script over
`git diff -U0 origin/main`: every changed line in a .cs file is blank or a
// / /// comment line. Build is 0 warnings / 0 errors with CS1591 still an
error; full unit test suite passes (5000 passed, 5 skipped).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs: give shared cursor-page and view-name properties real one-liners (#180 ) 2cc0a42d9c

Continues Pass F (XML-doc pass). Fixes the remaining public docs flagged as
pure restatements of their declaration (e.g. "The actors.", "The name."):
cursor-paginated response wire properties now say they're this page's items,
and DTO-specific properties (list/starter-pack names, singular refs) get a
one-liner naming what's distinctive about them instead of echoing the type.

Comment-only change; verified the same way as the previous commit. Build is
0 warnings / 0 errors, full unit test suite passes (5000 passed, 5 skipped).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs: dedupe the chat embed record doc onto its app.bsky original (#180 )

CI / pds-integration (pull_request) Successful in 31s

Details

CI / build-and-test (pull_request) Successful in 1m0s

Details

482087f689

Last piece of Pass F's item 3 (duplicated rationale outside Pass D's
files): chat.bsky.embed.record wraps the exact same EmbeddedRecordView
union as app.bsky.embed.record, so MessageRecordEmbedView.Record now
points at RecordEmbedView.Record with <inheritdoc cref> instead of
repeating its sentence. The stream-contract and auth/replay-store
rationale Passes A and B already centralized (EventStreamLoop /
StreamConsumerOptions, IJtiReplayStore) was checked and left alone.
Everything else the search turned up (per-type Unknown* union
boilerplate, sibling-method <param> docs, coincidentally similar
short phrases) is either already minimal or not a genuine duplicate,
so it was left as is.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

opencode-bot commented

2026-09-27 16:56:03 +00:00

🤖 Reviewing this PR — the review will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the review runs.

🤖 Reviewing this PR — the review will link the full session transcript.  📜 [Live transcript](https://git.grandiras.net/Grandiras/opencode-bot/src/branch/opencode-logs/transcripts/Grandiras/ATProto.NET/pr-184/run-2026-09-27T16-56-02-353Z.md) — updating every ~5 minutes while the review runs.

opencode-bot commented

2026-09-27 16:56:05 +00:00

⏳ Held on the OpenCode Go window — the 5-hour window is at 0% (the bot holds at 95%).
This continues automatically after the window resets (2026-09-27T21:56:02.846Z) — nothing to re-trigger.

⏳ **Held on the OpenCode Go window** — the 5-hour window is at 0% (the bot holds at 95%). This continues automatically after the window resets (2026-09-27T21:56:02.846Z) — nothing to re-trigger.

Grandiras commented

2026-09-27 16:56:08 +00:00

Comment-only check, run by the orchestrator over git diff -U0 origin/main...HEAD -- '*.cs': 5,591 changed lines, 0 outside comments or blank lines. No non-.cs files changed. Release build: 0 warnings, with CS1591 still an error and satisfied. 5,000 unit tests pass.

@claude /review, please. The diff is large but touches comments only (checked above), so please don't read all of it. Spot-check a sample of files for three things:

Spec, security or concurrency rationale that was lost when an XML doc on an internal member was deleted or turned into //. Look at Repo/, Streaming/, Auth/ and Server/ internals.
Rewritten public one-liners that are now wrong or misleading. The 53 name-restating Lexicon property docs were rewritten, for example "This page's actors." on the cursor-page wire twins.
<inheritdoc> or <see cref> targets that resolve to the wrong member.

Comment-only check, run by the orchestrator over `git diff -U0 origin/main...HEAD -- '*.cs'`: 5,591 changed lines, **0** outside comments or blank lines. No non-.cs files changed. Release build: 0 warnings, with CS1591 still an error and satisfied. 5,000 unit tests pass. @claude /review, please. The diff is large but touches comments only (checked above), so please don't read all of it. Spot-check a sample of files for three things: 1. Spec, security or concurrency rationale that was lost when an XML doc on an internal member was deleted or turned into `//`. Look at `Repo/`, `Streaming/`, `Auth/` and `Server/` internals. 2. Rewritten public one-liners that are now wrong or misleading. The 53 name-restating Lexicon property docs were rewritten, for example "This page's actors." on the cursor-page wire twins. 3. `<inheritdoc>` or `<see cref>` targets that resolve to the wrong member.

claude-bot commented

2026-09-27 16:56:12 +00:00

🤖 Reviewing this PR — follow along in the action run; the reply will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the task runs.

🤖 Reviewing this PR — follow along in [the action run](https://git.grandiras.net/Grandiras/claude-bot/actions/runs/639); the reply will link the full session transcript.  📜 [Live transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-184/run-639.md) — updating every ~5 minutes while the task runs.

claude-reviewer reviewed

2026-09-27 16:57:20 +00:00

claude-reviewer left a comment

💬 Verdict: comment — notes only, nothing blocking.

This is a large, mechanical, comment/doc-only pass, and the diff shown is consistent with that claim: every hunk touches only /// XML doc blocks being either collapsed to one line (public Lexicon members, not shown in this truncated diff) or converted to plain // comments (internal/private members). The internal/private detection logic described (tracking brace depth and container kind) appears correctly applied to the samples visible here — e.g. KeyedLock<TKey>.Lease.Dispose()/DisposeAsync() and SessionManager.BeforeSendAsync/TryRecoverAsync correctly lose their bare <inheritdoc/> because the enclosing types are internal, so CS1591 never required docs there.

One small inconsistency: in src/ATProtoNet/Auth/OAuth/DPoP.cs, the conversion strips <see cref> and <c> tags into plain prose everywhere, but leaves <see href="...">...</see> tags un-stripped in two places, producing raw pseudo-XML inside a plain // comment (worse than the original doc, which at least rendered as a link, and inconsistent with the PR's own stated tag-handling rules). Not a functional issue, just a leftover from the conversion script not handling the href variant of <see>.

Nothing else in the visible diff looks like an accidental code change, a misattributed doc comment, or a case where a doc that should have been public documentation got stripped as if internal. Given this is verified comment-only (per the PR's own script and loc.py), the CHANGELOG exemption for comment-only changes applies correctly.

⏱ 0m 53s · 1 turns · tokens: 72.9k in / 4.9k out (+23.9k cache-read)
📜 Full session transcript

💬 **Verdict: comment** — notes only, nothing blocking. This is a large, mechanical, comment/doc-only pass, and the diff shown is consistent with that claim: every hunk touches only `///` XML doc blocks being either collapsed to one line (public Lexicon members, not shown in this truncated diff) or converted to plain `//` comments (internal/private members). The internal/private detection logic described (tracking brace depth and container kind) appears correctly applied to the samples visible here — e.g. `KeyedLock<TKey>.Lease.Dispose()`/`DisposeAsync()` and `SessionManager.BeforeSendAsync`/`TryRecoverAsync` correctly lose their bare `<inheritdoc/>` because the enclosing types are internal, so CS1591 never required docs there. One small inconsistency: in `src/ATProtoNet/Auth/OAuth/DPoP.cs`, the conversion strips `<see cref>` and `<c>` tags into plain prose everywhere, but leaves `<see href="...">...</see>` tags un-stripped in two places, producing raw pseudo-XML inside a plain `//` comment (worse than the original doc, which at least rendered as a link, and inconsistent with the PR's own stated tag-handling rules). Not a functional issue, just a leftover from the conversion script not handling the `href` variant of `<see>`. Nothing else in the visible diff looks like an accidental code change, a misattributed doc comment, or a case where a doc that should have been public documentation got stripped as if internal. Given this is verified comment-only (per the PR's own script and `loc.py`), the CHANGELOG exemption for comment-only changes applies correctly. --- ⏱ 0m 53s · 1 turns · tokens: 72.9k in / 4.9k out (+23.9k cache-read) 📜 [Full session transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-184/run-639.md)

src/ATProtoNet/Auth/OAuth/DPoP.cs

					
				@ -6,31 +6,27 @@ using System.Text.Json;

				namespace ATProtoNet.Auth.OAuth;

The <see href="...">RFC 9449</see> tag is left intact inside a plain // comment instead of being unwrapped to plain text, unlike <see cref>/<c> elsewhere in this same conversion pass (e.g. the Thumbprint doc just below strips <c>crv</c> etc. but leaves its own <see href="...">RFC 7638</see> untouched too). Minor, but produces a comment with stray markup rather than the plain prose the rest of the pass aims for.

The `<see href="...">RFC 9449</see>` tag is left intact inside a plain `//` comment instead of being unwrapped to plain text, unlike `<see cref>`/`<c>` elsewhere in this same conversion pass (e.g. the `Thumbprint` doc just below strips `<c>crv</c>` etc. but leaves its own `<see href="...">RFC 7638</see>` untouched too). Minor, but produces a comment with stray markup rather than the plain prose the rest of the pass aims for.

Grandiras commented

2026-09-27 16:58:37 +00:00

@claude /review, please re-review. A new commit fixes your note: the two <see href="…">RFC …</see> tags left in plain // comments in DPoP.cs now read "RFC 9449 (https://…)". A repo-wide grep finds no other XML markup (<see, <c>, <paramref) in // comments, and the change is still comment-only.

@claude /review, please re-review. A new commit fixes your note: the two `<see href="…">RFC …</see>` tags left in plain `//` comments in `DPoP.cs` now read "RFC 9449 (https://…)". A repo-wide grep finds no other XML markup (`<see`, `<c>`, `<paramref`) in `//` comments, and the change is still comment-only.

Grandiras added 1 commit

2026-09-27 16:58:39 +00:00

docs(src): unwrap the RFC links left as markup in converted comments (#180 )

CI / pds-integration (pull_request) Successful in 29s

Details

CI / build-and-test (pull_request) Successful in 1m1s

Details

Sync Closures to GitHub / sync-closure (pull_request) Successful in 9s

Details

7f8d686ba9

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

claude-bot commented

2026-09-27 16:58:40 +00:00

🤖 Reviewing this PR — follow along in the action run; the reply will link the full session transcript.

📜 Live transcript — updating every ~5 minutes while the task runs.

🤖 Reviewing this PR — follow along in [the action run](https://git.grandiras.net/Grandiras/claude-bot/actions/runs/640); the reply will link the full session transcript.  📜 [Live transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-184/run-640.md) — updating every ~5 minutes while the task runs.

claude-reviewer approved these changes

2026-09-27 16:59:47 +00:00

claude-reviewer left a comment

✅ Verdict: approve — no blocking issues found.

Re-review after new commits. The one issue I flagged in my previous pass — <see href="...">RFC 9449</see> / <see href="...">RFC 7638</see> left as raw unstripped markup inside plain // comments in src/ATProtoNet/Auth/OAuth/DPoP.cs — is fixed by commit 7f8d686 ("unwrap the RFC links left as markup in converted comments"): both now read as plain prose with the URL in parentheses, e.g. RFC 9449 (https://www.rfc-editor.org/rfc/rfc9449), consistent with how <see cref>/<c> are handled elsewhere in the same pass.

The rest of the diff continues the same mechanical, comment/doc-only pattern as before: multi-line <summary>/<remarks>/<param> XML doc blocks on internal/private members (including public members of internal enclosing types, e.g. AtProtoPdsContainerResourceBase.Hostname, KeyedLock<TKey>.Lease, SessionManager) are converted to plain // prose, with <param>/<returns>/<exception> folded into labeled lines. Bare <inheritdoc/> is correctly dropped only where the enclosing type is internal (KeyedLock<TKey>.Lease.Dispose/DisposeAsync, SessionManager.BeforeSendAsync/TryRecoverAsync), matching the stated brace-depth/container-kind scanner logic. I didn't spot any place where genuinely public API documentation got miscategorized as internal, or where a doc's meaning changed during the conversion.

No remaining issues from my prior pass; nothing new stood out as a correctness or content problem in this batch.

⏱ 0m 54s · 12 turns · tokens: 80.2k in / 5.1k out (+1133.7k cache-read)
📜 Full session transcript

✅ **Verdict: approve** — no blocking issues found. Re-review after new commits. The one issue I flagged in my previous pass — `<see href="...">RFC 9449</see>` / `<see href="...">RFC 7638</see>` left as raw unstripped markup inside plain `//` comments in `src/ATProtoNet/Auth/OAuth/DPoP.cs` — is fixed by commit `7f8d686` ("unwrap the RFC links left as markup in converted comments"): both now read as plain prose with the URL in parentheses, e.g. `RFC 9449 (https://www.rfc-editor.org/rfc/rfc9449)`, consistent with how `<see cref>`/`<c>` are handled elsewhere in the same pass. The rest of the diff continues the same mechanical, comment/doc-only pattern as before: multi-line `<summary>`/`<remarks>`/`<param>` XML doc blocks on internal/private members (including public members of internal enclosing types, e.g. `AtProtoPdsContainerResourceBase.Hostname`, `KeyedLock<TKey>.Lease`, `SessionManager`) are converted to plain `//` prose, with `<param>`/`<returns>`/`<exception>` folded into labeled lines. Bare `<inheritdoc/>` is correctly dropped only where the enclosing type is internal (`KeyedLock<TKey>.Lease.Dispose/DisposeAsync`, `SessionManager.BeforeSendAsync/TryRecoverAsync`), matching the stated brace-depth/container-kind scanner logic. I didn't spot any place where genuinely public API documentation got miscategorized as internal, or where a doc's meaning changed during the conversion. No remaining issues from my prior pass; nothing new stood out as a correctness or content problem in this batch. --- ⏱ 0m 54s · 12 turns · tokens: 80.2k in / 5.1k out (+1133.7k cache-read) 📜 [Full session transcript](https://git.grandiras.net/Grandiras/claude-bot/src/branch/claude-logs/transcripts/Grandiras/ATProto.NET/pr-184/run-640.md)

Grandiras merged commit 199b7d1481 into main

2026-09-27 17:00:27 +00:00

Grandiras deleted branch debloat-f-docs

2026-09-27 17:00:27 +00:00

Grandiras referenced this pull request from a commit

2026-09-27 17:00:28 +00:00

docs(src): one-line XML docs, no docs on internal members (#184)

Grandiras referenced this pull request

2026-09-27 17:00:33 +00:00

refactor!: pre-release debloat pass #180

Grandiras referenced this pull request

2026-09-27 21:27:32 +00:00