Scan every Confluence space type, not just global (#191)

[richardmhope]

Closes #191. The last of the codebase-review issues.

The taxonomy, established

The issue said this needed verification against a real Cloud site. It didn't — every atlassian.com host is blocked by this environment's egress policy, but the spec reaches us another way. confluence.js v3.2.0 (README: "derived from Atlassian's OpenAPI spec") encodes the GET /spaces query parameter with Atlassian's own docstrings intact:

/** Filter the results to spaces based on their type. */
type: openEnum([
    'global', 'collaboration', 'knowledge_base', 'personal',
    'system', 'onboarding', 'xflow_sample_space',
]).optional(),

Corroborated by a web search for the distinctive string "xflow_sample_space" returning Atlassian's own space API docs as the top hit, and by CONFCLOUD-80579 "Add Collaboration and Knowledge_base Space types".

Seven types, and the filter takes one value — not a list. So type=global is not a narrowing, it is an omission of five, and there was never a server-side way to express "everything except personal".

Why this is worse than the issue estimated

Two of the five dropped types are collaboration and knowledge_base — the first two options Confluence's own space-creation flow offers. On any site created in recent years this isn't an edge case, it's most of the site. A knowledge base full of credentials was never read, and the scan completed looking clean over a scope the operator believed was covered.

The fix

Discovery lists every type and decides for itself:

• A personal space is excluded unless opted in — compared case-insensitively. The spec documents lower_case; there's a report that the v2 API returns UPPER_CASE.`` Trusting the documented spelling would scan every user's drafts on a site that asked not to — the same failure from the other side, so it's tested with three spellings.
• A space named explicitly in spaces is scanned whatever its type. Naming it is the deliberate decision the flag otherwise stands in for, and the alternative was reporting a space that plainly exists as not found.
• An unrecognised type is scanned, and logged. A type Atlassian adds after this release must not become content nobody looks at — excluding the unknown reopens the same hole from the other direction. But "is the new type personal-like?" is a question an operator should answer before a coverage report raises it for them, so confluence_unknown_space_types names it.

Accepted cost: personal spaces dominate the space count on a large site, so discovery now pages past them to discard each one — a few extra metadata requests once per scan. Against silently missing every space of an unnamed type, that's the right side of the trade, and the docstring says so.

Why the suite didn't catch it

confluence_server.py modelled only global and personal. It models the rest now, with a comment saying that's why the fixture exists.

Nine new tests; seven fail against the previous code (verified by stashing the connector and re-running). The two that pass either way guard the behaviour being preserved — personal spaces still excluded, and now case-insensitively.

make check green: ruff, mypy (285 files), docs check, 2035 passed / 2 skipped.

Operator action

Re-scan any Confluence source. Findings in non-global spaces have never been seen, so this is new coverage rather than a correction — until that scan runs, the source's history describes only part of the site. Noted in the CHANGELOG.

---

Generated by Claude Code

[icebergai-review-bot]

Verdict

APPROVE

Completed bounded review across 1 immutable scope(s). No actionable defects found in the supplied change.

Scope health

Convergence: healthy. Review mode: initial.
Recommended action: CONTINUE_INCREMENTAL.

• No escalation signals.

Prior findings

Finding	Status
—	No prior finding state

New findings

No new findings.

Fix-induced regressions

• None evidenced.

Uncertainty

• No material uncertainty recorded.

Validation

• Reviewed the supplied immutable patch and its added Confluence discovery coverage.

Residual risks

• None identified.