PDF Processing changedagents-and-tools/agent-skills/best-practices
Nearest release: v2.1.233, published under an hour before this site recorded the change. Shown because the two are within 24 hours of each other. Nothing here says the release caused the edit.
Recorded here
Lines+34added
Lines−28removed
From line
270
where the diff opens
First seen
14 Aug 2026
this site's first read of the page
Recorded edits6to this page, all time
The whole hunk
from line 270, old and new numbered
/
from line 270
270270
271271The complete Skill directory structure might look like this:
272272
273```text
274pdf/
275├── SKILL.md # Main instructions (loaded when triggered)
276├── FORMS.md # Form-filling guide (loaded as needed)
277├── reference.md # API reference (loaded as needed)
278├── examples.md # Usage examples (loaded as needed)
279└── scripts/
280 ├── analyze_form.py # Utility script (executed, not loaded)
281 ├── fill_form.py # Form filling script
282 └── validate.py # Validation script
283```
273* `pdf/`
284274
275 * `SKILL.md`: Main instructions (loaded when triggered)
276
277 * `FORMS.md`: Form-filling guide (loaded as needed)
278
279 * `reference.md`: API reference (loaded as needed)
280
281 * `examples.md`: Usage examples (loaded as needed)
282
283 * `scripts/`
284
285 * `analyze_form.py`: Utility script (executed, not loaded)
286 * `fill_form.py`: Form filling script
287 * `validate.py`: Validation script
288
285289#### Pattern 1: High-level guide with references
286290
287291````markdown
from line 318
314318
315319For Skills with multiple domains, organize content by domain to avoid loading irrelevant context. When a user asks about sales metrics, Claude only needs to read sales-related schemas, not finance or marketing data. This keeps token usage low and context focused.
316320
317```text
318bigquery-skill/
319├── SKILL.md (overview and navigation)
320└── reference/
321 ├── finance.md (revenue, billing metrics)
322 ├── sales.md (opportunities, pipeline)
323 ├── product.md (API usage, features)
324 └── marketing.md (campaigns, attribution)
325```
321* `bigquery-skill/`
326322
323 * `SKILL.md` (overview and navigation)
324
325 * `reference/`
326
327 * `finance.md` (revenue, billing metrics)
328 * `sales.md` (opportunities, pipeline)
329 * `product.md` (API usage, features)
330 * `marketing.md` (campaigns, attribution)
331
327332````markdown SKILL.md
328333# BigQuery Data Analysis
329334
from line 1066
10611066
10621067**Example:**
10631068
1064```text
1065bigquery-skill/
1066├── SKILL.md (overview, points to reference files)
1067└── reference/
1068 ├── finance.md (revenue metrics)
1069 ├── sales.md (pipeline data)
1070 └── product.md (usage analytics)
1071```
1069* `bigquery-skill/`
1070
1071 * `SKILL.md` (overview, points to reference files)
1072
1073 * `reference/`
1074
1075 * `finance.md` (revenue metrics)
1076 * `sales.md` (pipeline data)
1077 * `product.md` (usage analytics)
10721078
10731079When the user asks about revenue, Claude reads SKILL.md, sees the reference to `reference/finance.md`, and calls bash to read just that file. The sales.md and product.md files remain on the filesystem, consuming zero context tokens until needed. This filesystem-based model is what enables progressive disclosure. Claude can navigate and selectively load exactly what each task requires.
10741080
No line in this hunk matches that.