The first code block of "For --update" in references/update.md (unchanged
through v0.9.72, line 17) both persists and prints the incremental detection
result:
result = detect_incremental(Path('INPUT_PATH'))
new_total = result.get('new_total', 0)
print(json.dumps(result, indent=2, ensure_ascii=False))
Path('graphify-out/.graphify_incremental.json').write_text(json.dumps(result, ensure_ascii=False), encoding="utf-8")The result carries every file path in the corpus (files) plus the changed
subset (new_files). On a ~940-file corpus that print was roughly a thousand
indented paths.
Why it matters in this skill specifically: the reader of that stdout is not a
person scrolling a terminal, it is the agent's context window. Every later step
already reads .graphify_incremental.json from disk, so the print duplicates
the most verbose artifact of the pipeline into the scarcest resource, at the
start of a flow that still has to dispatch subagents and run the merge.
It also contradicts the full build. Step 2 of SKILL.md handles the
equivalent detect() result with an explicit "Do NOT cat or print the JSON",
followed by a short per-category summary. The incremental path is the only one
that dumps it.
Observed vs expected: observed, a full JSON dump of every path. Expected,
only what the next decision depends on, with the data left on disk, the same
as Step 2.
Proposal: replace the dump with a summary, keeping the sidecar write as is:
nf = result.get('new_files', {})
print(json.dumps({
'new_total': new_total,
'deleted': len(result.get('deleted_files', [])),
'new_by_type': {t: len(v) for t, v in nf.items()},
'total_files': result.get('total_files'),
'total_words': result.get('total_words'),
'skipped_sensitive': result.get('skipped_sensitive', []),
}, indent=2, ensure_ascii=False))
print('Full result: graphify-out/.graphify_incremental.json')I ran the whole --update flow with this substitution and every later step
behaved identically, since none of them read stdout.
The same pattern may be worth a pass elsewhere in the generated skills: any
block that writes a sidecar and also prints it in full.
Environment: observed on graphifyy 0.9.55, graphify install --platform claude,
Claude Code on Windows 11; now on 0.9.65. Block checked unchanged in v0.9.72.
The first code block of "For --update" in
references/update.md(unchangedthrough v0.9.72, line 17) both persists and prints the incremental detection
result:
The result carries every file path in the corpus (
files) plus the changedsubset (
new_files). On a ~940-file corpus that print was roughly a thousandindented paths.
Why it matters in this skill specifically: the reader of that stdout is not a
person scrolling a terminal, it is the agent's context window. Every later step
already reads
.graphify_incremental.jsonfrom disk, so the print duplicatesthe most verbose artifact of the pipeline into the scarcest resource, at the
start of a flow that still has to dispatch subagents and run the merge.
It also contradicts the full build. Step 2 of
SKILL.mdhandles theequivalent
detect()result with an explicit "Do NOT cat or print the JSON",followed by a short per-category summary. The incremental path is the only one
that dumps it.
Observed vs expected: observed, a full JSON dump of every path. Expected,
only what the next decision depends on, with the data left on disk, the same
as Step 2.
Proposal: replace the dump with a summary, keeping the sidecar write as is:
I ran the whole
--updateflow with this substitution and every later stepbehaved identically, since none of them read stdout.
The same pattern may be worth a pass elsewhere in the generated skills: any
block that writes a sidecar and also prints it in full.
Environment: observed on graphifyy 0.9.55,
graphify install --platform claude,Claude Code on Windows 11; now on 0.9.65. Block checked unchanged in v0.9.72.