v1.1.0

Same 61 skills as v1.0.0, corrected. Every numerical example in this release has
been executed; twenty commits of review fixed what that turned up, and CI now
runs the examples on every push so it cannot silently rot again.

Corrected examples. A dollar-bar accumulator that never reset, so four
$800 trades produced $1,600/$800/$800 bars. A rolling Sharpe inflated by the
square root of its own window. Drift-detection breakpoints that discarded the
tails they exist to measure. A purge bound that ignored the end of the training
set. A quarter bucket built on datetime64[Q], which is not a numpy unit and
raised on every input, so the block had never run. Meta-labels fitted on primary
predictions made in sample, and a walk-forward PCA fitted across folds. A kill
switch that ran its breach handler and then let the breaching order through
against a book it had just flattened. A monitoring loop that defined a daily
loss limit and never read it, and paged on a single sample.

Installer. --prune drops skills this repo no longer ships; --uninstall
removes the install. Both delete only what the installer created. --copy is
re-runnable, switching between symlinks and copies works in both directions, and
a skill you copied in by hand is never replaced.

CI. A skill-examples job executes the ### CORRECT blocks straight out of
the Markdown. Frontmatter is parsed with PyYAML rather than a hand-rolled
splitter that accepted malformed YAML and rejected valid forms. The API check
now verifies call signatures, not just names, and reports the library versions
it checked against. The README catalog must list every skill.

v1.0.0

61 agent skills covering the Machine Learning for Trading workflow: data
sourcing and point-in-time correctness, feature engineering and labelling,
purged and combinatorial cross-validation, cost-aware backtesting, portfolio
risk, live trading, and autonomous research operators.

Each skill is a single SKILL.md file pairing a wrong pattern with the correct
one, then handing off to the published ml4t-* libraries where a stable API
exists. No executable scripts, no hooks, no network calls.

Install. Skill discovery is one level deep, so the category directories in
this archive have to be flattened. Unpack it and run ./scripts/install.sh,
which links or copies each skill in as ml4t-<skill-name>.

Checked in CI. Structure and frontmatter, the 120-line limit, the
concept-first rule, an acyclic dependency graph, a regenerated chapter map, and
every ml4t.* name in a Production Implementation snippet against the packages
published on PyPI.