docs: refresh the GLPI agent skills to 0.4.1

[baraline]

skills/ teaches an agent how to call this package. Nothing imported it, no
build read it, and neither mypy nor ruff nor Sphinx looked at it. It had rotted
three ways at once, and none of them broke anything anyone would notice.

What was wrong

Six of the seven published skills did not load. The published copies opened
with a UTF-8 BOM. Frontmatter is recognised by the file starting with ---,
so three invisible bytes cost each skill its name and description — and the
description is the field an agent reads to decide whether a skill is relevant at
all. Only glpi-client-setup, the one file without a BOM, ever reached a
session.

Twenty-five public methods were named by no skill anywhere — the whole
knowledge base family (18) and the whole Fields-plugin family (7). The skills
declared version: 0.3.0/0.4.0; the package is at 0.4.1.

Thirty-two claims in the remaining seven were false or stale, and the
routing agent's error contract described a different library: requests
exceptions (requests is not a dependency — the package uses httpx),
tenacity.RetryError (which cannot escape, the transport sets reraise=True),
and bare ValueError for 4xx instead of the GlpiStatusError hierarchy.

What changed

• A coverage tripwire — every public client method must be named as a whole
  word by some skill, with an allowlist that starts empty. This is what would
  have caught the original gap: every existing check validated what the skills
  said, and none asked what they omitted.
• Two new skills. glpi-knowledge-base (18 methods) carries the gotcha that
  categories cannot be written through v2 at all — they go through the legacy v1
  API and raise a plain RuntimeError, not a GlpiError, and a failed create is
  not rolled back. glpi-plugin-fields (7 methods) documents that the Fields
  plugin is absent from the v2 contract entirely, so every method needs a v1
  session, and that values has two different shapes depending on which helper
  you call.
• The seven older skills corrected, including a from_env snippet that did
  not compile (async with at module level), glpi-ticket-workflow claiming
  "there is no batch iterator" when iter_search_tickets exists, and
  get_ticket_statistics silently truncating at one page of 200.
• The agent's error contract rewritten to the hierarchy the package actually
  raises, including the three plain RuntimeError sites that are not GlpiError.
• An absolute BOM guard in the marketplace drift checker. The existing
  BOM ONLY verdict was comparative, so a BOM faithfully copied into both
  trees read as IN SYNC — which is exactly the state six skills were in while
  not loading at all.

Verification

• pytest -m "not integration" --cov --cov-fail-under=95 → 938 passed, 97.11%
• check_skill_drift.py → exit 0, 9/9 in sync, no BOM offenders
• BOM gate proven non-vacuous by planting probes in three plugin directories
  (exit 1, all named) and removing them (exit 0)
• All nine skills at 0.4.1; zero BOMs in either tree; every python block parses
• Every method, model and enum member named in a skill verified against
  inspect.signature and __all__

Known follow-ups (not blocking)

• check_skill_drift.py's _plugin_dirs() returns [] when no plugin.json is
  found and main() has no floor check, so the BOM sweep could go vacuous while
  still exiting 0. Harmless today; one line to fix.
• Separately, and outside this branch: get_ticket_statistics reads one page of
  200 with no pagination (_statistics.py:402-405), yet CHANGELOG.md:59-60
  attributes a live 16,245-vs-1,552 ticket measurement to it. Those cannot both
  be true.

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.10%. Comparing base (e6ed643) to head (2d09875).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #21   +/-   ##
=======================================
  Coverage   97.10%   97.10%           
=======================================
  Files          79       79           
  Lines        2422     2422           
=======================================
  Hits         2352     2352           
  Misses         70       70           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.