Customer-readable description
What this skill can do
Research a limited post window of a public X account: first collect, then retrieve by scope and conditions, and finally output a summary, categorized list, or downstream processing input. For known accounts, use the script; for unknown accounts, first do candidate discovery, then enter account research.
How customers use it
First confirm four items: account/candidate bloggers, time or latest post count, keyword/topic/evidence conditions, and artifact type. By default, only produce a research summary; do not automatically generate images. Convert natural-language phrases like "the last week/month" into explicit CST dates first.
Common script entry points:
python3 scripts/profile_x.py https://x.com/<handle> \
--limit <N> --since <YYYY-MM-DD> --until <YYYY-MM-DD> \
--query <term> --output-mode summary --output <run-dir>
Repeat --query defaults to OR; to require both, add --query-mode all;--topic is an alias. For full parameters and output fields, see Research Contract; for reusable commands, see Query Recipes。
Dependencies and installation
- Python 3 standard library; network mode accesses
api.fxtwitter.com; offline mode uses--source-json。 - Does not read X cookies, login state, or private content.
- image is not a default dependency; only when explicitly selecting
image-prompts/alldoes it enter the downstream image skill.
Progressive execution
- L0 Scope: confirm account/candidates, window, filter conditions, and artifacts; if information is insufficient, echo assumptions first.
- L1 Collection: limited pagination, first
--dry-run, record fetched, pagination, and stop reason. - L2 Research: write
summary.md、filtered.json、classification.yml, preserving post URL, time, and evidence. - L3 Downstream: only write when the customer explicitly needs it
image-prompts.yml/prompts/, then hand off tosoia-media-generate-article-imagefor bitmap and visual acceptance.
Routing principles
- Account summary: select only
summary, do not compile a Prompt. - Time/topic retrieval: use the combination of
--since/--until、--query/--topic、--category、--has-media/--has-alt. - Model/prompt research: output research results first; use only when explicit downstream conversion is needed
--output-mode image-prompts。 - Unknown blogger discovery: first use X search or Web search to find candidate accounts/posts, record sources and reasons for inclusion, then run this skill on the candidate accounts; this script does not pretend to search the entire web.
Boundaries and deliverables
latest-window does not equal the full account history;all is also limited by --max-pages . Prompt Deck is not bitmap acceptance, GPT2 is only a source model clue, not a new preset. For output directories, evidence retention, failure receipts, and image import fields, see Research Contract and soia-open-media-content-skills/skills/soia-media-generate-article-image/references/prompt-x-profile-import.md。
Forward tests
For three real scenarios and acceptance criteria, see three-scenario forward testing;xiaoxiaodong01 is only one of the regression samples, not a default account.
Logs and completion receipts
Each page outputs page / received / added / total. The receipt must include account/candidates, scope, conditions, fetched, selected, classification, GPT2/ALT, coverage completeness, pagination limit, and artifact path; "Prompt compiled" must not be written as "image generation passed".
Private information and intermediate data
The run bundle is placed in the temporary or formal output directory specified by the client, not in the repository, vault root, or ~/.codex/generated_images. The public repository only commits scripts, contracts, and offline tests; it does not commit real account scraping results, ALT, provider caches, or generated images.
Install the entire domain (Claude Code and Codex share the same domain plugin):
claude plugin marketplace add soia-team/soia-open-skills
claude plugin install soia-pkm-vault@soia
Install only this one skill:
npx skills add soia-team/soia-open-pkm-vault-skills -g -a '*' -s soia-pkm-clip-x-profile -y
WorkBuddy 's loading unit is role-based experts rather than plugins,npx skills add -a '*' cannot cover it; it needs to be installed separately, see docs/install/workbuddy.md。