Cv Update Review

CV Update Review

Overview

Run a public, profile-driven CV update workflow. The skill discovers candidate activities, drafts a review packet, requires explicit user approval, then edits only a dated copy of the Word CV.

This skill is intentionally confirm-first. It must never silently decide that an item belongs on the CV.

Public-First Design

Do not hard-code private file paths, employer details, author identities, or CV section names in this skill. Use a local config file instead.

Default config path:

.cv-update-review/config.json

Configs may use ${BASECAMP_ROOT} for vault-relative paths, especially when an iCloud/Obsidian vault lives under different usernames on different machines. The helper scripts resolve ${BASECAMP_ROOT} from the environment when set, or from the nearest vault containing START_HERE.md and AGENTS.md.

If no config exists, run setup:

python3 .agents/skills/cv-update-review/scripts/setup_config.py \
  --output .cv-update-review/config.json

Use profiles/example.json as a public-safe example only. Do not edit it with real personal data.

Eligibility Rule

Default eligibility is strict:

• include only completed, public, or accepted activities
• include PubMed-indexed or accepted publications
• include public/accepted lectures, invited talks, presentations, public media,

publicly released digital scholarship, and completed service/leadership items

• exclude planned, drafted, submitted-but-not-accepted, internal-only, or

speculative work

• exclude routine recurring items unless the profile marks them as milestones

For podcasts, newsletters, websites, or recurring educational media, default to major milestones rather than every routine release unless the user explicitly changes the rule.

See references/eligibility-rules.md before resolving borderline items.

Monthly Workflow

1) Resolve the Window

Use the system-local date for relative dates.

Use review_schedule.cadence from the config when present.

Default monthly window:

• start: first day of the prior completed calendar month
• end: last day of the prior completed calendar month

Quarterly window:

• start: first day of the prior completed calendar quarter
• end: last day of the prior completed calendar quarter

Ad hoc only:

• ask the user for the target month or date range before scanning

Custom cadence:

• follow review_schedule.custom_cadence if it is concrete enough to resolve a

date window; otherwise ask the user before scanning

If the user names a month or date range, use that exact range.

2) Load Configuration

Read .cv-update-review/config.json or the user-supplied config path.

Required fields:

• cv_document
• output_dir
• at least one evidence source:
• pubmed.author_queries
• rss_feeds
• website_urls
• local_activity_paths
• local_search_roots
• manual_scholar_sources

If required fields are missing, run setup or ask for the missing values before scanning.

If review_schedule.offer_automation is true and no automation exists yet, offer to configure one. A CV automation must run in packet_only_until_user_approval mode: it may scan sources and prepare the review packet, but it must not create or edit a Word CV copy until the user approves exact entries, sections, wording, and evidence sufficiency.

3) Gather Evidence

Use the smallest source set that can answer the request.

Local sources:

• configured activity logs
• configured project/status files
• configured journal or worklog paths
• configured folders containing releases, talks, manuscripts, abstracts, or

teaching artifacts

Optional helper:

python3 .agents/skills/cv-update-review/scripts/scan_local_activity.py \
  --config .cv-update-review/config.json \
  --start YYYY-MM-DD \
  --end YYYY-MM-DD \
  --output .cv-update-review/evidence/local_YYYY-MM.json

Public sources:

• PubMed through NCBI E-utilities:

python3 .agents/skills/cv-update-review/scripts/scan_pubmed.py \
  --config .cv-update-review/config.json \
  --start YYYY-MM-DD \
  --end YYYY-MM-DD \
  --output .cv-update-review/evidence/pubmed_YYYY-MM.json

• configured RSS/Atom feeds:

python3 .agents/skills/cv-update-review/scripts/scan_rss.py \
  --config .cv-update-review/config.json \
  --start YYYY-MM-DD \
  --end YYYY-MM-DD \
  --output .cv-update-review/evidence/rss_YYYY-MM.json

• Google Scholar/profile checks:
• use configured profile/search URLs as a manual or browser verification

source

• do not build a fragile Scholar scraper into the required workflow
• if a colleague configures a permitted third-party provider, record that as a

profile-specific source and cite it in the packet

Collaboration and mentoring sources:

• scan configured project/status files for collaborator names, trainee roles,

student/fellow/resident labels, observerships, mentoring language, and supervised deliverables

• identify whether a project involved a junior collaborator or mentee who may

belong under Mentoring Activities even if the project itself is not yet a completed CV work

• keep mentoring candidates separate from project-output candidates; do not

treat an unfinished project as CV-ready just because it reveals a mentoring relationship

4) Show Candidate Preview

Before asking for private recall, show the user the actual candidate items found by source. Do not provide only counts.

For each source, include:

• short candidate label/title
• activity type and proposed or provisional CV section
• evidence pointer: URL, file path and line number, PMID, or feed name
• eligibility status

Keep the preview concise enough to scan. Group clear duplicates or repeated project-status heartbeat lines when needed, but preserve enough evidence detail for the user to decide what is missing or wrongly included.

5) Category And Section Cross-Check

Before finalizing the packet, classify each candidate by activity type:

• manuscript or journal publication
• abstract or poster
• talk, invited lecture, course/session lecture, or conference presentation
• teaching aid, public media, website, software, or educational artifact
• mentoring, supervision, observership, or trainee collaboration
• service, committee, leadership, review, grant, or award

Use the activity type to choose the CV section. Titles alone are not enough. When two items share a similar title, keep them separate if their activity types differ. For example, a conference talk and a journal article with similar titles must be evaluated and placed independently: the talk belongs with talks or conference presentations, while the journal article belongs with manuscripts or published research articles.

If the same publication list appears in more than one CV location, inspect both locations and decide whether both need the new item to preserve the existing CV structure. Record that decision in the packet.

6) Collaboration/Mentee Review

Before the recall prompt, ask whether configured collaboration projects include junior collaborators, trainees, students, residents, fellows, visiting scholars, or observers who should be considered under Mentoring Activities.

Show any names and roles already found from project/status evidence. For each, include:

• name
• inferred role or level
• project or activity source
• proposed mentoring wording
• confidence and what still needs confirmation

Do not insert mentoring entries until the user confirms the person, role, institution, supervision years, and wording.

7) Prompt For Private Recall

Before finalizing the packet, ask:

What completed, public, or accepted professional activities from  may be missing from public records or local files? Examples: accepted manuscripts, invited talks, delivered lectures, committee/service work, peer review, mentorship milestones, grants/awards, public media, or completed institutional activities.

If the user provides items, ask for enough evidence to classify them:

• date
• activity type
• title/name
• role
• venue/institution/platform
• public URL, acceptance notice, program, email, certificate, or local file path

Do not include private recall items in the confirmed update list until the user supplies sufficient evidence or explicitly accepts a low-evidence packet note.

8) Build The Review Packet

Create a dated Markdown packet before any Word edit:

python3 .agents/skills/cv-update-review/scripts/build_update_packet.py \
  --config .cv-update-review/config.json \
  --start YYYY-MM-DD \
  --end YYYY-MM-DD \
  --evidence .cv-update-review/evidence/local_YYYY-MM.json \
  --evidence .cv-update-review/evidence/pubmed_YYYY-MM.json \
  --evidence .cv-update-review/evidence/rss_YYYY-MM.json \
  --output /cv-update-packet_YYYY-MM-DD.md

Use the packet shape in references/packet-template.md.

Each candidate must show:

• activity type
• proposed CV section
• draft wording
• evidence
• confidence
• eligibility status
• unresolved questions

9) Stop For Approval

Stop after the packet unless the user has already approved specific entries in the current conversation.

Require approval at entry level:

• include/exclude
• final CV section
• final wording
• whether evidence is sufficient

Do not treat "looks good" as approval if the packet still contains unresolved section or wording questions.

10) Extract CV Style Before Editing

Before editing a Word CV, inspect the current document's conventions near the target sections:

python3 .agents/skills/cv-update-review/scripts/extract_cv_style.py \
  --docx "/path/to/cv.docx" \
  --section "Published and Accepted Research Articles" \
  --section "Formal Teaching"

Match the existing CV, including:

• author order and punctuation
• journal abbreviations, years, volume/issue/pages, DOI style
• abstract database wording
• lecture title quotation style
• venue/date/location order
• mentorship table or list format
• capitalization quirks already present in that section

Do not globally normalize the CV unless the user explicitly asks.

11) Create A Dated Word Copy

Never edit the original CV directly.

Create a dated copy:

python3 .agents/skills/cv-update-review/scripts/copy_dated_cv.py \
  --config .cv-update-review/config.json \
  --date YYYY-MM-DD

Then edit the dated copy only. Prefer the dedicated document-editing tools when available. If editing the .docx package directly, follow references/docx-editing-rules.md.

12) Verify The Updated Copy

After editing:

• extract text from the dated copy and confirm approved entries appear exactly
• confirm excluded entries do not appear
• confirm original CV remains unchanged
• render or visually inspect the Word/PDF output when possible
• return a concise changelog with the dated copy path

Output Contract

After a discovery run, return:

• packet path
• source window
• number of candidates by eligibility status
• items requiring user confirmation
• whether no Word document was edited

After an approved Word update, return:

• dated CV copy path
• entries inserted, grouped by section
• verification performed
• any residual uncertainty

Failure Handling

• If config is missing, run setup or ask for setup values.
• If PubMed is unavailable, continue with local/RSS/manual sources and mark

PubMed unavailable in the packet.

• If RSS parsing fails for one feed, continue with other feeds and record the

failed feed.

• If Google Scholar cannot be checked safely, leave it as a manual verification

task in the packet.

• If the CV document is missing, stop before packet application and report the

missing path.

• If a dated copy already exists, create a suffix such as _v2 unless the user

explicitly authorizes overwrite.

• If visual verification cannot run, report the next best text extraction check.

References

• references/public-setup-guide.md: setup flow for colleagues
• references/eligibility-rules.md: include/exclude decisions
• references/packet-template.md: review packet structure
• references/docx-editing-rules.md: Word copy and edit safety rules

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: erafat
• Source: erafat/skills
• License: MIT

Install and usage instructions live in the source repository linked above.