feat(changelog): one-line summaries by default, --full to expand

[padak]

What

kbagent changelog now shows a one-line summary per version by default instead of a wall of text. --full (-v) expands every note as before.

Before (0.57.0 was a single ~3 KB bullet):

v0.57.0
  • BREAKING (flow / conditional flows): the flow command group now targets conditional flows
    (keboola.flow) ONLY; keboola.orchestrator support is dropped. --component-id is removed from
    every flow subcommand and from the /flows REST surface (FlowCreate/FlowUpdate/FlowSchedule
    ... [~40 more lines] ...

After:

v0.57.0
  • BREAKING (flow / conditional flows): the flow command group now targets conditional flows
    (keboola.flow) ONLY; keboola.orchestrator support is dropped.  (+4 more)

Run with --full (-v) to see complete notes.

Why

The changelog data had drifted from its own contract ("one-line summaries per version") into multi-paragraph essays, and the renderer — though it word-wraps and colours prefixes — can't rescue a single 3 KB bullet. Two compounding problems: data (mega-paragraphs) and rendering (no summary view). This fixes both.

Changes

Rendering (commands/changelog.py)

• Default = summary: first note's headline + dim (+N more) + a --full footer hint when detail is hidden.
• --full / -v: full notes, word-wrapped (previous behaviour).
• New recognised prefix BREAKING: (bold red) in _PREFIX_STYLES + _PREFIX_RE — previously it got no prefix styling.
• --full bound via functools.partial so the --json payload stays {entries} only (flag is presentation, not data).

Summarisation (changelog.py)

• New headline(): first sentence, capped at CHANGELOG_HEADLINE_MAX_CHARS on a word boundary, with guards so periods inside version numbers (0.57.0) and abbreviations (e.g.) aren't read as sentence ends, plus a dangling-backtick cleanup.
• format_whats_new (post-update banner) now summarises the same way.

Data + authoring

• The 0.57.0 entry is re-split from one mega-paragraph into five prefixed bullets as the reference example.
• Authoring contract (one logical change per bullet, recognised prefix, self-contained first sentence) documented in the changelog.py module docstring and CONTRIBUTING.md.

Config: CHANGELOG_HEADLINE_MAX_CHARS = 160 in constants.py (no hardcoded literal in the renderer).

Docs synced: CLAUDE.md command list, AGENT_CONTEXT (context.py), commands-reference.md.

Tests

• tests/test_changelog_render.py (16 cases): headline() guards (version-number, abbreviation, truncation, dangling backtick), summary vs --full, (+N more), BREAKING styling, clean JSON payload, format_whats_new summarisation.
• Full non-e2e suite green (3884 passed).

Notes for reviewers

• No version bump. This intentionally ships in a later release PR — that PR must add the changelog entry for this feature (none is added here, since 0.57.0 is already tagged).
• The 0.57.0 entry re-split is editorial (same information, reorganised) and stands on its own.
• make typecheck baseline (47 pre-existing diagnostics) is unchanged; none touch the changed files.

🤖 Generated with Claude Code

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 4 additional findings.

[padak]

Devin review findings — addressed in e1c1249

The 4 Devin findings live in the Devin app (not posted as GitHub inline threads), so there's nothing to resolve here — recording the disposition for reviewers:

#	Finding	Disposition
1	changelog.py:38 — 0.57.0 "Change (flow validation)" bullet has a ~600-char first sentence, doesn't honour the new "self-contained first sentence" contract	Fixed. Split into a tight 149-char first sentence so the reference example actually models the contract.
2	commands/changelog.py:168 — partial() binding compatible with OutputFormatter.output	No action — verification only; confirmed correct (and covered by test_json_payload_has_no_full_key).
3	changelog.py:859 — a sentence genuinely ending in a version number (0.56.0 entry) isn't split	No action — inherent, ambiguous heuristic limitation; the truncation fallback keeps it scannable and the authoring contract steers new entries away from it. Devin flagged it as not-a-bug.
4	changelog.py:860 — the before.isdigit() guard also suppresses !/? after a digit (e.g. "exit code 5!")	Fixed. Scoped the guard to periods only (note[dot] == "."); a digit before !/? is now a genuine boundary. Added a regression test.