Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/workflows/testarch/bmad-teach-me-testing/SKILL.md`:
- Around line 12-13: The invocation mapping in SKILL.md declares pause/resume
capability but lacks explicit resume keywords; update the mapping that sets mode
to "create" (the line starting "If user invoked with 'create' or 'teach' or
'learn' or 'start'") to include "resume" and "continue" as triggers, and make
the same addition to the duplicated mapping block referenced at lines 104-117 so
the menu/routing matches the documented pause-and-resume behavior.

In `@src/workflows/testarch/bmad-testarch-automate/SKILL.md`:
- Around line 63-67: The doc currently presents a "tri-modal step-file
architecture" but also lists a separate Resume option, causing ambiguity; update
the SKILL.md wording so Resume is described explicitly as a continuation of the
Create flow (not a fourth mode) — for example change the Resume menu entry to "-
[R] Resume — Resume an interrupted Create workflow" and adjust the surrounding
description where "tri-modal step-file architecture" and the menu entries for
Create/Validate/Edit (the Create mode (steps-c/), Validate mode (steps-v/), Edit
mode (steps-e/), and the Resume entry) are documented (also apply the same
change around lines referenced in the second occurrence) so the doc consistently
frames Resume as a Create-flow continuation rather than a standalone mode.

In `@src/workflows/testarch/bmad-testarch-automate/steps-e/step-02-apply-edit.md`:
- Around line 62-66: The terminal hook invocation using "python3
.../resolve_customization.py --skill {skill-root} --key workflow.on_complete"
lacks a failure fallback; update the step documentation for
resolve_customization.py (the call with --key workflow.on_complete) to state
that if the script fails you must manually resolve workflow.on_complete by
checking, in order, {skill-root}/customize.toml,
{project-root}/_bmad/custom/{skill-name}.toml, and
{project-root}/_bmad/custom/{skill-name}.user.toml, and only if the resolved
workflow.on_complete is non-empty follow it as the final terminal instruction
before exiting.

In `@src/workflows/testarch/bmad-testarch-ci/SKILL.md`:
- Around line 63-67: The SKILL.md wording is ambiguous about "Resume" suggesting
both tri-modal and four-mode behavior; update the mode model entries (the
tri-modal list around the "Create mode (steps-c/), Validate mode (steps-v/),
Edit mode (steps-e/)" block and the related bullet list around the Resume entry)
to explicitly state that "[R] Resume — Resume an interrupted workflow
(Create-mode continuation)" so Resume is clearly a Create-flow path, and ensure
any other mentions in the nearby section (previously lines ~75-84) are changed
consistently to reflect Resume = Create continuation.

In `@src/workflows/testarch/bmad-testarch-ci/steps-e/step-02-apply-edit.md`:
- Around line 64-66: The current terminal step assumes running
resolve_customization.py --key workflow.on_complete always succeeds; update the
step to handle resolver failure by checking the script's exit status or output
and treating failures as non-fatal: if the resolver returns a successful result
and workflow.on_complete is non-empty, execute it as the final terminal
instruction; if the resolver fails (non-zero exit, missing output, or parse
error), skip executing workflow.on_complete, log or print a clear warning about
the resolver failure, and exit normally; reference the resolve_customization.py
invocation and the workflow.on_complete key when implementing this fallback.

In `@src/workflows/testarch/bmad-testarch-framework/SKILL.md`:
- Around line 63-67: Update the tri-modal wording in SKILL.md so Resume is
framed as part of Create mode: modify the Create mode bullet (the one mentioning
"Create mode (steps-c/)") to note that the resume path is included (e.g.,
reference step-01b-resume.md) and make the same change in the corresponding
block around lines 75-84 to ensure consistency between the tri-modal description
and the Resume routing; update any text that currently lists Resume as a
standalone choice so it is described as a Create-mode branch under steps-c/ and
step-01b-resume.md.

In `@src/workflows/testarch/bmad-testarch-nfr/customize.toml`:
- Around line 33-35: The persistent_facts glob persistent_facts =
["file:{project-root}/**/project-context.md"] is too broad and can match
multiple project-context.md files leading to undefined behavior; either narrow
it to a single explicit path (e.g., change the value to
file:{project-root}/project-context.md) or, if multiple files must be supported,
update the loading logic that consumes persistent_facts to implement and
document an explicit strategy (e.g., "first match wins" or "merge all") and add
tests to cover multiple-match behavior so the behavior of persistent_facts is
deterministic.

In `@src/workflows/testarch/bmad-testarch-test-design/SKILL.md`:
- Around line 63-67: The documentation currently describes both "tri-modal" and
a separate "Resume" mode, causing a contradiction; update the SKILL.md text
where Create/Validate/Edit modes are listed (the "Create mode (steps-c/)" bullet
and similar mentions around lines ~75-84) to explicitly fold Resume behavior
into Create mode by noting "Create mode (steps-c/): primary execution flow (new
+ resume)" and remove any standalone "Resume" mode references so the taxonomy
reads consistently as three modes: Create (including resume), Validate
(steps-v/), and Edit (steps-e/).

In
`@src/workflows/testarch/bmad-testarch-test-review/steps-c/step-04-generate-report.md`:
- Around line 113-117: The "On Complete" step currently calls
resolve_customization.py for workflow.on_complete without fallback or clear
semantics; update the step to catch script errors and provide a deterministic
fallback: if running python3
{project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key
workflow.on_complete fails, skip the on_complete hook and proceed to exit, and
if the resolver returns a non-empty workflow.on_complete, explicitly state that
it should be executed as a shell command (not merely "followed") as the final
terminal instruction before exiting.

In `@src/workflows/testarch/bmad-testarch-trace/SKILL.md`:
- Around line 63-67: Update the tri-modal description in SKILL.md to explicitly
state whether "Resume" is a separate mode or an entrypoint of Create mode (e.g.,
change the Create mode bullet under the header so it reads that Create
(steps-c/) includes the resume entrypoint), and make the same clarification
within the later block referenced around lines 75-84; locate and edit the
bullets for "**Create mode (steps-c/)**", "**Validate mode (steps-v/)**", and
"**Edit mode (steps-e/)**" so they consistently indicate that "Resume" is an
entrypoint of Create rather than a fourth mode (or vice versa if you intend
Resume to be a distinct mode), ensuring the document no longer presents
contradictory mode semantics.

---

Outside diff comments:
In `@test/test-installation-components.js`:
- Around line 243-265: The test currently only validates SKILL.md for
teach-me-testing (teachMeSkillPath) but omits the
customize.toml/workflow.md-removal assertions run in the workflowDirs loop;
update the test so bmad-teach-me-testing is included in that loop by adding the
key 'bmad-teach-me-testing': 'teach-me-testing' to the workflowDirs object (and
remove the separate teachMeSkillPath block) OR copy the same customize.toml,
persistent_facts, and workflow.md negative assertions into the existing
teach-me-testing try/catch branch so the same checks run for teachMeSkillPath
and the workflowDirs iteration.

---

Duplicate comments:
In `@src/workflows/testarch/bmad-teach-me-testing/customize.toml`:
- Around line 33-35: The persistent_facts entry uses a recursive glob
"file:{project-root}/**/project-context.md" that can match multiple files and
cause undefined merges; replace it with a single, deterministic path or a more
specific glob so only one file is matched (e.g., point to the exact
project-context.md location or restrict the pattern to a single expected
directory) and update the persistent_facts entry (the string
"file:{project-root}/**/project-context.md" under persistent_facts) to that
deterministic path.

In `@src/workflows/testarch/bmad-teach-me-testing/steps-c/step-05-completion.md`:
- Around line 341-345: Add robust error handling around the call to
resolve_customization.py and make the "follow it" instruction explicit: when
running python3 _bmad/scripts/resolve_customization.py --skill {skill-root}
--key workflow.on_complete, capture and handle failures (non-zero exit,
exceptions, or empty output) and log the error, then fall back to the same
default/abort behavior used in SKILL.md activation steps; if the resolver
returns a non-empty string, treat that string as the exact terminal instruction
to execute (document which executor/command runs it), otherwise proceed with the
defined fallback path.

In `@src/workflows/testarch/bmad-teach-me-testing/steps-v/step-v-01-validate.md`:
- Around line 265-269: The "On Complete" step calls the resolver script (python3
_bmad/scripts/resolve_customization.py --skill {skill-root} --key
workflow.on_complete) but lacks error handling and a clear fallback and the
phrase "follow it as the final terminal instruction" is ambiguous; update the
step to run the resolver and if the script exits non‑zero or returns an empty
value, log or display a clear fallback instruction (matching SKILL.md's
activation pattern) and proceed with that fallback, and clarify that when a
non‑empty workflow.on_complete is returned you should execute those instructions
as the final step (or state the explicit fallback steps to take if execution
fails) so callers of resolve_customization.py and readers of the doc know both
success and error paths.

In
`@src/workflows/testarch/bmad-testarch-test-design/steps-v/step-01-validate.md`:
- Around line 69-73: The "On Complete" step invoking python3
_bmad/scripts/resolve_customization.py to resolve workflow.on_complete lacks
error handling and is ambiguous about execution semantics; update the step to
run the resolver guardedly (capture script non-zero exit and missing/empty
output) and provide a clear fallback/explicit error message and behavior (e.g.,
log error and continue or abort), and clarify whether the resolved
workflow.on_complete is to be executed as a shell command versus treated as AI
terminal instructions; reference the resolver script (resolve_customization.py)
and the resolved key (workflow.on_complete) and align behavior with the SKILL.md
activation pattern used in step-04-generate-report.md for consistent error
handling.

In `@src/workflows/testarch/bmad-testarch-test-review/customize.toml`:
- Around line 33-35: The persistent_facts entry uses a recursive glob
"file:{project-root}/**/project-context.md" which can match multiple files and
cause undefined merge behavior; update the persistent_facts configuration to
target a single, deterministic file or a stricter path (for example replace the
glob with an explicit path to the intended project-context.md or narrow the glob
to a single known folder) so only one file is matched; locate the
persistent_facts key in customize.toml and change the value referencing
"file:{project-root}/**/project-context.md" accordingly.

In
`@src/workflows/testarch/bmad-testarch-test-review/steps-v/step-01-validate.md`:
- Around line 69-73: The step's invocation of the resolver script (python3
_bmad/scripts/resolve_customization.py --skill {skill-root} --key
workflow.on_complete) lacks error handling and the phrase "follow it" is
ambiguous; update the step to (1) check the resolver process exit status and
capture stderr/stdout, (2) if the script fails or returns empty, fall back to a
clear default behavior (e.g., proceed with no-op or a documented default
terminal action) and log the error output, and (3) clarify that if
workflow.on_complete resolves to non-empty text the agent must treat that text
as the final terminal instruction to execute exactly once before exiting;
reference resolve_customization.py, the --key workflow.on_complete flag, and the
workflow.on_complete result when implementing these checks and the fallback.

---

Nitpick comments:
In `@src/workflows/testarch/bmad-testarch-atdd/customize.toml`:
- Around line 8-10: The header comment currently flags three merge rules
including "arrays-of-tables with `code`/`id`: replace matching items," which
doesn't apply to this `[workflow]` file (it only uses scalars and flat arrays);
remove or trim that third rule so the header only documents "scalars: override
wins" and the array-append behavior for persistent_facts/activation_steps_* to
avoid confusion, and apply the same edit to the matching headers in the other
variants (`bmad-testarch-test-design`, `bmad-testarch-framework`,
`bmad-testarch-trace`) to keep them consistent.

In `@src/workflows/testarch/bmad-testarch-nfr/SKILL.md`:
- Around line 40-42: Step 3's wording is ambiguous about how file: entries in
{workflow.persistent_facts} are resolved; update the SKILL.md text to clearly
state that entries prefixed with file: are interpreted as paths or globs that
may be absolute, may include the {project-root} placeholder (as in the
customize.toml default file:{project-root}/**/project-context.md), or may be
given as paths relative to {project-root} — resolve any relative paths under
{project-root} and expand globs accordingly. Apply the same clarification to the
ATDD and test-review SKILL.md copies so all three documents use the same
explicit phrasing about placeholder resolution and relative vs absolute paths.

In `@test/test-installation-components.js`:
- Around line 289-290: The negative contains checks on skillContent can falsely
pass if the file is empty; update the pathExists(skillMdPath) block so that
after reading skillContent you assert it has a sensible minimum size (e.g.,
assert(skillContent && skillContent.trim().length > 0, `SKILL.md for ${dirName}
should not be empty`) or a specific length threshold) before running the
negative assertions that reference skillContent.includes(...); place this guard
immediately after skillContent is loaded so functions/variables like
pathExists(skillMdPath) and skillContent are validated first.