Reduce the Atlassian skills to Jira, and fix eight broken modules

[johnml1135]

The two Atlassian skills supported Jira, Confluence and Bitbucket. FieldWorks uses Jira. Confluence and Bitbucket are gone — 4,362 deletions against 293 insertions — and eight modules that never worked are fixed.

Start here: .claude/skills/atlassian-skills/SKILL.md — the Configuration and Core Workflow sections show the shape of the whole change.

Where to look:

• Eight of twelve read-only modules were unusable. They raised NameError on import, so any read through them failed and callers had to reach into the write skill to do a read. Cause is an upstream generator bug: the read-only variant is produced by stripping write functions, and the stripper also pruned Optional and AtlassianCredentials from imports while surviving signatures still referenced them. Reported as langpingxue/atlassian-skills#14.
• Confluence and Bitbucket removed to the roots — not just the eight script modules per skill, but their docs, config blocks, credential fields, the CQL reference, and the plumbing in _common.py: dataclass fields, is_*_available, get_*_client, and the service branches in from_credentials. Zero references remain outside the provenance notes.
• What remained was rewritten, not left with holes. Configuration is two short modes with SIL Data Center first. Core Workflow, the agent-mode example and the credentials reference are Jira-shaped rather than two-thirds excised.
• Both frontmatter descriptions advertised Confluence and Bitbucket and are rewritten. An overstated description is how the wrong skill gets loaded.
• The read-only variant also lost its duplicated utility catalogue: 237 lines restating function signatures REFERENCE.md already documents. Now a generated module-to-function index.

Deliberately not here: _common.py keeps its check_available_skills machinery even though only one service can now be unavailable. It is load-bearing for the credentials path and cheap to leave.

Verification: all 9 modules in each skill import (8 were broken); a live jira_get_issue('LT-22723') and jira_get_transitions both succeed through the read-only skill — previously impossible; zero Confluence/Bitbucket references outside PROVENANCE.md. gitlint clean. No build.ps1/test.ps1: Python utilities outside the build, no compiled code.

Next: review and merge — independent of #1098/#1099/#1100/#1102, touching none of their files.

---

Size before and after

	Before	After
atlassian-skills (md + py)	~5,900	3,325
atlassian-readonly-skills (md + py)	~4,400	2,219
Script modules per skill	17	9
Broken modules (read-only)	8	0

SKILL.md specifically: write variant 741 → 382 lines, read-only 560 → 148.

What the licence investigation found

Both skills carry license: Complete terms in LICENSE in frontmatter, which is what stopped an earlier pass from touching them at all. Checking properly:

Question	Answer
Where from?	https://github.com/langpingxue/atlassian-skills, vendored in d1a9bc6
Declared licence?	MIT, in the upstream README
LICENSE file upstream?	No. GitHub's license endpoint 404s
Copyright line anywhere?	No
LICENSE file here?	No — the frontmatter reference dangles in both places

MIT permits modification, so the earlier caution was unnecessary. The attribution it does ask for was missing and now exists as PROVENANCE.md in each skill. Since upstream names no copyright holder, a verbatim MIT text could not honestly be authored on their behalf; the provenance files record the declaration and its gaps instead.

This is now a hard fork. A re-sync is no longer a merge — upstream becomes a source to cherry-pick Jira fixes from. PROVENANCE.md says so.

Data Center gotchas captured while investigating

Upstream's docstrings describe Jira Cloud; SIL's instance is Data Center. Three that silently disagree, now in atlassian-skills/PROVENANCE.md:

• assignee wants a username, not an accountId. jira_create_issue and jira_update_issue both send {"accountId": ...}, which Data Center rejects. Pass custom_fields={"assignee": {"name": "<username>"}}.
• Affects Version is not exposed at all. Pass custom_fields={"versions": [{"name": "FW 9.3"}]}.
• resolution cannot be set by an update — not on the edit screen; only a transition sets it.

Each was found by hitting it. They are duplicated into jira-issue/references/publish.md on #1098, where someone working an LT ticket looks first.

---

This change is 

[github-actions]

NUnit Tests

    1 files  ±0      1 suites  ±0   10m 35s ⏱️ -25s
5 853 tests ±0  5 772 ✅ ±0  81 💤 ±0  0 ❌ ±0 
5 862 runs  ±0  5 781 ✅ ±0  81 💤 ±0  0 ❌ ±0 

Results for commit f5c09c3. ± Comparison against base commit 6f72671.

♻️ This comment has been updated with latest results.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 38.34%. Comparing base (6f72671) to head (f5c09c3).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1101   +/-   ##
=======================================
  Coverage   38.33%   38.34%           
=======================================
  Files        1507     1507           
  Lines      350580   350580           
  Branches    40293    40293           
=======================================
+ Hits       134410   134426   +16     
+ Misses     186941   186925   -16     
  Partials    29229    29229           

see 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.