chore(refresh): refreshed configuration and documentation

[rios0rios0]

Automated weekly refresh by config-and-docs-refresh.yaml in rios0rios0/config-automation.

Claude Code reviewed the in-scope configuration and documentation files (currently CLAUDE.md, .github/copilot-instructions.md, and .github/skills/code-review/SKILL.md) against the current code, updated whichever drifted, and recorded the change as a .changes/unreleased/ fragment (chlog repositories) or under [Unreleased] in CHANGELOG.md (everywhere else).

Review the diff carefully --- this PR was generated by Claude Code and may contain inaccuracies.

[claude]

Claude finished @rios0rios0's task in 1m 20s —— View job

---

Review: chore(refresh): refreshed configuration and documentation

This PR is a one-line documentation correction plus its changelog fragment. The Feature Catalog line in CLAUDE.md ("Context the AI receives") still advertised a 32 KiB bound on the reviewed repository's CLAUDE.md and a 16 KiB bound on the PR description — figures that no longer match the code and that contradicted four other places in the same document. The PR restates them as 1 MiB and 64 KiB. No behaviour changes; nothing under internal/, cmd/ or .github/ is touched.

Verdict: approve. The corrected numbers are the right ones, and the change removes a real internal contradiction rather than introducing one.

Inline threads posted

None — no candidate finding reached the ≥80 confidence bar.

What I checked, per dimension

CLAUDE.md compliance (clean) — The repository's own mandate that every change ships a .changes/unreleased/ fragment is honoured: 1789371835880909921-47f7.yaml matches its siblings byte-for-byte in shape (quoted kind/body/time, RFC 3339 nanosecond timestamp), the body is simple past tense starting lowercase ("corrected the …"), CHANGELOG.md is correctly left untouched, and no --breaking flag is warranted for a documentation correction. README.md needs no update because no behaviour or configuration changed — only the description of an existing default.

Correctness (clean) — I verified both corrected figures against the source rather than against the document:

• internal/domain/entities/settings.go:254 — const defaultMaxGuidelinesBytes = 1024 * 1024 → 1 MiB, resolved by GuidelinesBytes() at :260-265 on an unset or non-positive value. Pinned by settings_test.go:543.
• internal/domain/entities/settings.go:275 — const defaultMaxPRDescriptionBytes = 64 * 1024 → 64 KiB, resolved by PRDescriptionBytes() at :280-285. Pinned by settings_test.go:555.

Both are wired to the command layer through MaxGuidelinesBytes / MaxPRDescriptionBytes at all three call sites (review_controller.go:143-144, review_all_controller.go:98-99, webhooks/dispatcher.go:368-369), so the defaults the document now names are the ones a deployment actually gets.

Internal consistency (clean) — after this change the four other statements of these bounds agree with the catalog line: the project-guidelines flow (1 MiB ≈ 256k tokens ≈ 25% of a 1M-token window), the PR-metadata flow (default 64 KiB), and the two configuration-table rows for ai.max_guidelines_bytes and ai.max_pr_description_bytes. The stale line was the last holdout.

Code-comment adherence (clean) — the doc comments on the two constants describe 1 MiB as "roughly 256k tokens" and 64 KiB as "roughly 16k tokens". That second phrase is the likely origin of the stale 16 KiB figure — a tokens-vs-bytes mix-up — and the new text states the byte bound, which is what the surrounding sentence is about.

Git history (clean) — git log/-S on the modified line shows no earlier commit that deliberately set 32 KiB/16 KiB as anything other than an initial figure left behind when the defaults were raised. This is not a regression of a recorded decision.

Security (clean) — the diff introduces no code, no credentials, no argv or log surface, and no downloaded or executed content. A Markdown line and a YAML changelog fragment.

Design (clean) — the change fits the document's established structure: the Feature Catalog states the bound tersely and defers the rationale to the detailed flow section, which is exactly how the other bounded inputs are documented.

Findings raised and dropped

Finding	Score	Reason dropped
design — fragment kind should be Fixed (it corrects an inaccuracy) rather than Changed	25	Both are defensible under Keep a Changelog for a documentation correction; the repository's own convention does not distinguish them for doc-only changes, and sibling fragments use Changed for comparable work. Not worth a thread.
design — the catalog line states 1 MiB without the "lower it on a small-window backend" caveat that accompanies the figure elsewhere	20	The caveat lives in the project-guidelines flow section and the configuration table, which is where the document consistently puts rationale. The catalog is deliberately terse.
correctness — 1 MiB guidelines default could overflow a small-window backend, making the documented default misleading	15	Pre-existing design decision, explicitly documented and deliberate. Out of scope: this PR describes the default, it does not set it.
· branch chore/config-and-docs-refresh

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[Copilot]

🔵 Needs a closer look

Two unresolved documentation corrections remain.

Pull request overview

Refreshes documentation to reflect current prompt-budget defaults and records the update.

Changes:

• Updated CLAUDE.md.
• Added an unreleased change fragment.

File summaries

File	Summary
CLAUDE.md	Documents 1 MiB and 64 KiB default budgets; wording and a contradictory docs/COMPARISON.md entry still need correction.
.changes/unreleased/1789371835880909921-47f7.yaml	Records the documentation refresh.

Review details

Suppressed comments (2)

CLAUDE.md:34

• These are configurable budgets, not fixed bounds: GuidelinesBytes() and PRDescriptionBytes() honor any positive ai.max_*_bytes value and use 1 MiB/64 KiB only as defaults. Describing them simply as “1 MiB bound” and “64 KiB description bound” makes this catalog inaccurate for configured deployments; call them default budgets or name the settings.

**Context the AI receives** — the diff (fenced per file with language tag); operator rules (Markdown files, YAML frontmatter `paths` globs, universal + language-matched categories); the reviewed repository's own root `CLAUDE.md` (any provider, 1 MiB bound, skipped when the PR modifies it); **PR intent metadata** — title and source/target branch names (prompt header) plus description and commit count (fetched via the `prmetadata` registry from the GitHub/ADO REST APIs, 64 KiB description bound) with explicit "judge intent / flag scope creep" guidance; prior bot review threads with all replies (mention re-review path only).

CLAUDE.md:34

• This refresh still leaves docs/COMPARISON.md:230 describing the shipped PR-description budget as 16 KiB, while AIConfig.PRDescriptionBytes() defaults to 64 KiB. Since this line updates the same feature's documented bounds, please update that comparison entry in the same refresh so the repository documentation does not contradict itself (the documentation-refresh rule in .github/skills/code-review/SKILL.md:229 requires this).

**Context the AI receives** — the diff (fenced per file with language tag); operator rules (Markdown files, YAML frontmatter `paths` globs, universal + language-matched categories); the reviewed repository's own root `CLAUDE.md` (any provider, 1 MiB bound, skipped when the PR modifies it); **PR intent metadata** — title and source/target branch names (prompt header) plus description and commit count (fetched via the `prmetadata` registry from the GitHub/ADO REST APIs, 64 KiB description bound) with explicit "judge intent / flag scope creep" guidance; prior bot review threads with all replies (mention re-review path only).

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.