Move compile-time variable substitution out of the LLM and into a
deterministic Python step. SKILL.md becomes a two-line stdout-dispatch
shim that runs render.py and follows the instruction it prints. The
renderer reads BMad configuration from the central four-layer TOML
surface introduced in #2285 (_bmad/config.toml plus config.user.toml
and the two _bmad/custom/ overrides), with a fallback to the legacy
per-module _bmad/bmm/config.yaml for pre-#2285 installs.
Compile-time refs ({{.var}}) get substituted at render time. LLM-runtime
refs ({var}) pass through untouched.
Renderer (render.py)
- Python 3 stdlib only (tomllib, already bundled since 3.11). UTF-8 I/O.
Every invocation rebuilds from scratch — no hash, no cache.
- find_project_root walks up from cwd; HALT to stdout if no _bmad/
is found anywhere on the path.
- load_central_config deep-merges the four TOML layers in priority
order (base-team → base-user → custom-team → custom-user) so user
overrides in _bmad/custom/config.user.toml win over installer-
regenerated base values. flatten_central_config lifts scalar keys
from [core] and [modules.bmm] into the renderer's flat namespace;
module keys beat core on collision (matches the installer's own
core-key-stripping behavior).
- When _bmad/config.toml is absent, falls through to the legacy
flat-YAML parser for _bmad/bmm/config.yaml — the renderer keeps
working across the #2285 transition.
- {{.var}} substitution; unresolved refs emit empty string (Go
missingkey=zero semantics).
- Smart defaults for planning_artifacts / implementation_artifacts /
communication_language applied after config load. Derives
sprint_status / deferred_work_file from implementation_artifacts.
{{.main_config}} points at whichever surface was actually read.
- Renders every .md in the skill dir except SKILL.md to
{project-root}/_bmad/render/bmad-quick-dev/.
- On success, stderr summary plus a single stdout line:
"read and follow {workflow_md}". On failure, stdout HALT directive —
per the Anthropic skills spec, script stdout is the defined agent-
communication channel.
Skill entry (SKILL.md)
- Two-line shim: run python render.py, follow stdout. No template
tokens in SKILL.md itself.
Template conversions
- workflow.md, step-01..05, step-oneshot, sync-sprint-status: convert
every compile-time {var} reference to {{.var}}. Runtime refs
preserved.
- spec-template.md untouched (single-curly comment hint stays as
documentation).
Skill-prose cleanups bundled in
- Remove dead step-file frontmatter: empty-string variable declarations
(spec_file, story_key, diff_output, review_mode) in quick-dev step-01
and code-review step-01; empty --- --- blocks in step-03 and step-05;
the specLoopIteration counter init moved from step-04 frontmatter into
the step body where first-entry vs loopback semantics are explicit.
- Unify the language rule across all six quick-dev step files plus
workflow.md.
Tooling
- tools/validate-skills.js: add TPL-01 rule. Files whose name contains
"template" must not contain compile-time {{.var}} substitutions.
Template files seed durable, version-controlled artifacts that
execute on other machines; baking a value at render time would
freeze a machine-local path into every downstream artifact.
- tools/validate-file-refs.js: add render/ to INSTALL_ONLY_PATHS so
the validator recognizes the runtime-generated buffer.
- tools/skill-validator.md: document TPL-01; deterministic rule count
bumped from 14 to 15.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* refactor(bmad-ux): replace bmad-create-ux-design with lean spine-based bmad-ux
* refactor(bmad-ux): adopt DESIGN.md spec, split into two-file spine, align prd/brief
DESIGN.md (visual identity per the Google Labs spec) and EXPERIENCE.md
(behavior, flow, IA) replace the single design.md spine. EXPERIENCE.md
cross-references DESIGN.md tokens via the spec's {path.to.token} syntax.
Example suite restructure
- 3 DESIGN.md examples: editorial (Stitch source / Linen & Logic), calm
native mobile (Quill), shadcn-on-Tailwind web SaaS (Drift)
- 2 paired EXPERIENCE.md examples (Quill, Drift); Linen & Logic unpaired
to model the Stitch handoff scenario
- Replaces the prior 2-example combined spine set
Discovery additions (outcome-driven, one line each)
- Source scan: glob {planning_artifacts}/ for candidates, parent never reads
- Form-factor: resolve before IA closes; journeys often derive it
- Surface closure: every stated need has a surface, every surface a journey
- Named-protagonist journeys (Mary, not "the user")
- Design handoff working mode (extensible producer registry, default: Stitch)
PRD and brief alignment with same insights
- bmad-prd: dropped standalone Primary Persona section from template;
renamed "Personas + Journeys" entry to "Journey-led"; named-protagonist
rule on UJs; form-factor probe; validation checklist updated
- bmad-product-brief: form-factor surfaced in Discovery topics
Quality scan fixes
- Added ## Overview heading; renamed ## Activation to ## On Activation
- Replaced ../ paths in example assets with {planning_artifacts}/
- Sources section compressed (abstract delta-only rule)
- Working mode aligned to "Fast path" / "Coaching path" BMad-wide convention
New
- references/design-md-spec.md: working summary of the spec for the LLM
- customize.toml: design_md_examples, experience_md_examples,
design_handoffs registries
- .prettierignore: ignore .analysis/ quality-scan artifacts repo-wide
* refactor(bmad-ux): activation parity with prd/brief, opt-in reviewer gate, no headline grade
- Restructure On Activation as numbered six-step list mirroring bmad-prd
and bmad-product-brief, restoring the explicit key-resolution list that
earlier crammed-paragraph form had dropped (planning_artifacts and
friends were silently unresolved at Create).
- Make Reviewer Gate opt-in and lens-selectable. At Finalize, ask before
spending tokens on parallel reviewer subagents; at Validate intent,
skip that question but still confirm lens picks. Stops the auto-run
WCAG audit on hobby-stakes work.
- Drop the overall validation grade. Per-category verdicts and severity
counts already say what is true; a single headline grade conflated
design rigor with release readiness and led "POOR" pills landing on
reports whose own bodies described the work as strong. Removed from
references/validate.md (ladder rule + markdown twin), HTML template
(grade pill div + CSS vars + classes).
- Trim creative-tools.md: drop the Custom entries section. Runtime
prompt files should only carry what the LLM needs to act in this
moment; how-to-extend-via-TOML is setup-time human documentation
already covered by customize.toml comments.
* fix(bmad-ux): align validation report template with 8-category rubric
Template placeholders referenced 'Decision-readiness' and 'seven dimensions'
from the prior rubric. Replace with TEMPLATE_CATEGORY_NAME and inline the
eight canonical categories from references/validate.md so the synthesis pass
names them verbatim.
* fix(validate-skills): remove stale WF-01/WF-02 rules
WF-01/WF-02 were originally scoped to workflow.md files (now mostly gone)
but had been generalized to flag name/description in any non-SKILL.md
markdown. That over-captured legitimate spec files — e.g. DESIGN.md
examples in bmad-ux/assets/ that carry name/description per the Google
Labs DESIGN.md spec.
Step files are already covered by STEP-06. Rule count: 14 → 12.
* fix(bmad-ux): address PR review followups
- validation-report-template.html: severity badge class is badge-sev-*,
not sev-* (the comment misled the synthesis pass).
- Sweep dangling bmad-create-ux-design references: module-help.csv,
bmad-agent-ux-designer/customize.toml, bmad-prd/SKILL.md handoff list,
workflow-map.md (en + 4 translations), getting-started.md (en + 4
translations). Workflow-map output column updated to DESIGN.md +
EXPERIENCE.md.
- references/validate.md: Markdown capitalized as a proper noun.
- SKILL-04: require bmad- prefix, enforce single dashes via regex
^bmad-[a-z0-9]+(-[a-z0-9]+)*$, drop FORBIDDEN_NAME_SUBSTRINGS
- WF-01/WF-02: check all .md files (not just workflow.md) for stray
name/description frontmatter, with tech-writer exception
- Update skill-validator.md prompt to match all rule changes
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add tools/validate-skills.js — a Node CLI that checks 13 deterministic
rules (SKILL-01–06, WF-01–02, PATH-02, STEP-01/06/07, SEQ-02) across
all skill directories. Runs in under a second, exits non-zero on HIGH+
findings in strict mode, and outputs JSON for the inference validator.
- Add validate:skills npm script to quality chain
- Update skill-validator.md with first-pass integration instructions
- Update AGENTS.md push gate documentation
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace duplicated workflow-era rules in .augment/code_review_guidelines.yaml
with a single reference to tools/skill-validator.md. Append the skill spec
cheatsheet to the validator with a link to the Agent Skills specification.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace `[workflow.md](workflow.md)` with bare `workflow.md` in all 34
SKILL.md files. Redundant markdown link syntax adds noise for LLM
consumers. Also update the validator example to match.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Run skill-validator against all 36 skills on HEAD (42b1d0f6). Fixes:
- PATH-01: relative path corrections in product-brief-preview, brainstorming,
distillator, generate-project-context, edit-prd, quick-dev-new-preview
- PATH-04: remove 5 intra-skill path variables from edit-prd
- REF-01: single-curly template vars → double-curly in create-architecture,
create-epics-and-stories, create-story
- REF-02: fix dangling file refs in advanced-elicitation, validate-prd, edit-prd
- REF-03: update rule to prefer natural language `Invoke the skill` form;
fix stale persona ref in qa-generate-e2e-tests
- SEQ-01: "skip to" → "proceed to" in quick-dev-new-preview
- SEQ-02: remove time estimates from document-project, quick-spec
- SKILL-06: add "Use when" trigger to review-edge-case-hunter
- STEP numbering: renumber step n="0.5" to integer sequence in retrospective
Also updates REF-03 rule in tools/skill-validator.md to clarify that
natural language invocation is canonical — no skill: prefix required.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Skills must be invoked with "invoke" language, not file-oriented verbs
like "read fully and follow", "execute", "run", or "load". These imply
document-level operations and are incorrect for skill references.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add PATH-05: no file path references into another skill directory.
Skills are encapsulated — external consumers must use skill:name syntax,
not reach into internal files. Also tighten WF-03 to cross-reference
PATH-05 so vague "legitimate external path" no longer permits
cross-skill file paths.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
SKILL-05 checks that SKILL.md name matches the directory name.
REF-01 checks that every {variable} traces to frontmatter, config,
or runtime — exempts {{double-curly}} template placeholders.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Alex Verkhovsky
Brian