Skip to content

v1.2.0

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 19 Apr 06:04
v1.2.0
33552f8

Persona-add scaling refactor. Reduces the contributor surface for
adding a new persona from ~85 hand-written touchpoints to ~7 source
files, and introduces a featured/catalog tier so the landing demos
don't grow unboundedly.

No behavior change for the five installed personas.

Schema additions to rules//_meta.json

  • featured (bool): whether the persona appears in landing-page demos.
    All five current personas set featured: true. Personas #6+ default
    to featured: false (catalog tier) — they appear in tables and
    dropdowns but not in hand-written demo prose. load_meta enforces
    strict-bool typing.
  • register_short (string): one-line summary used by table generators
    across surfaces. Single source vs the previous three different
    per-surface descriptions. load_meta enforces non-empty and rejects
    table-breaking characters (|, newlines, backticks).

scripts/render.py extensions (~750 lines added)

  • Marker-based partial injection: render_zone, inject_marker,
    check_marker, find_markers. Multi-format markers (HTML comments
    for .md/.mdc, # comments for .sh/.yml/.yaml). Marker ids enforced
    as [a-z0-9-]+ across all four functions. Recursion guard refuses
    content containing the same marker id.

  • JSON field-replace: update_json_field / check_json_field. Compares
    parsed-data semantic equality (not text-equality), so hand-edited
    4-space-indented JSON is NOT rewritten by no-op mutators. Accepts
    both mutate-in-place and return-new-dict mutator styles.

  • Featured/catalog filters with strict-bool semantics.

  • Schema validation in load_meta: required fields, exactly one default
    flavor, non-empty/safe register_short, non-empty triggers,
    strict-bool featured. Catches contributor typos at the
    canonical-source boundary.

  • Eight pure-function partial generators (gen_persona_table,
    gen_personas_command_table, gen_slash_commands_table,
    gen_subagents_table, gen_trigger_phrases_list, gen_cp_skills_block,
    gen_hooks_allow_list, gen_dramaturg_skills,
    gen_personas_catalog_index_body). Empty input raises ValueError.
    Trigger text is markdown-escaped to handle special characters.

  • Two new full-file outputs: commands/personas.md and agents/dramaturg.md.
    Body templates extracted to scripts/templates/ — keeps prose edits
    in plain markdown. commands/personas.md fixes the v1.1.0 stale
    4-row table bug (was missing ontario-bud).

  • All Path.read_text / write_text now pass encoding="utf-8" (was
    inheriting platform default — broken on Windows cp1252).

  • Consumer-file marker injection (inject_consumer_markers /
    check_consumer_markers). Always loads the FULL persona set,
    ignoring the caller's subset, so scripts/render.py shakespeare
    cannot silently corrupt README/docs/hooks tables. Per-zone
    errors are batched into one ValueError so contributors see all
    failing zones at once. Distinguishes "marker pair missing"
    (structural) from "content drifted" in error messages.

Marker-injected zones in consumer files

13 zones across README.md, docs/index.md, docs/personas/index.md,
docs/subagents.md, docs/install/{claude-code,index}.md,
hooks/{activate,session-start}.sh. Content between markers is
auto-rendered from rules//_meta.json; surrounding prose
stays hand-edited. CI gate (--check) detects manual edits inside
zones and fails the PR.

README trim

The per-persona Demo blocks (~80 lines of hand-curated flavor demos
for 4 of 5 personas, never updated for ontario-bud) are replaced
with one short example + link to the docs site. Removes ongoing
per-flavor demo maintenance.

Tests: 51 → 124 (+73)

New tests/test_partial_injection.py covers marker mechanism, JSON
field-replace, featured filter, load_meta validation (11 rejection
cases), empty-input guards, markdown escaping, recursion guard,
consumer-marker wiring (subset-render protection, missing-vs-drift
errors, batched failures), and registry validity (every registered
marker exists in its file, no duplicates).

tests/test_render.py adds TestHandWrittenPerPersonaFiles enforcing
that every persona has commands/.md and agents/-reviewer.md
with the right frontmatter — prevents the v1.1.0 bug class
(persona PR merging without these files) from recurring.

tests/test_manifest.py and tests/test_render.py replace two
hardcoded 4-tuples with dynamic enumeration of rules/. The hardcoded
lists were missing ontario-bud — masking real coverage gaps that
shipped in v1.1.0.

Documentation

CONTRIBUTING.md "Adding a new persona" section rewritten with the
new ~7-file workflow plus an explicit featured-tier promotion
process (maintainer-only, capped at 5).

CLAUDE.md replaces the two-class generated/hand-written split with
the new three-class taxonomy (fully generated / marker-injected
zones / hand-written) and points to CONTRIBUTING.md for the
add-a-persona checklist.

Commits since v1.1.0

  • fix(render): Phase 2 review hardening — subset corruption, register_short validation, missing-marker error, batched inject
  • chore: release v1.2.0
  • refactor(render): wire marker injection into consumer files; trim README; rewrite CONTRIBUTING
  • refactor(render): marker-injection foundation + featured tier + auto-gen commands/personas + dramaturg
Assets 4
Loading