Job application

Create truthful, role-specific application documents from the bundled structured profile or a profile supplied by the user.

Privacy and authorization

• Treat all profile and application data as private.
• Never invent qualifications, employment, education, or contact details.
• The bundled profile is fictional demonstration data and may be replaced by the user.
• Do not submit, email, upload, or message an application without explicit authorization for that specific destination.

Persistent data and inputs

• Profile data: $JOB_APPLICATION_PROFILE_DIR/*.json when configured, otherwise the fictional demonstration data in assets/profile/*.json
• Application output: $JOB_APPLICATION_OUTPUT_DIR when configured, otherwise $JOB_APPLICATION_DATA_DIR/applications (or the platform data directory's equivalent)
• Job data: a job.json shaped like examples/job.json
• Templates: assets/templates/

Keep real profiles, photographs, signatures, generated documents, and application history outside the installed skill directory. ClawHub updates may replace that directory. Resolve bundled resources relative to this skill directory, but honor the persistent-data environment variables. Do not assume a username, home directory, workspace layout, messaging channel, or recipient ID. For ClawHub, export the whitelist-only staging directory described in PUBLISHING.md, then run the publish dry-run against that staging directory.

Workflow

1. Create the application folder with python3 scripts/prepare_application.py --slug <yyyy-mm-company-role>. For a derived application, add --base-application <absolute/base-directory>; its readable, valid cv_selection.json is checked before a new directory is created. Use the printed absolute path for all following files. Never overwrite an existing application directory unless the user explicitly requests reuse.

2. Save normalized job details as job.json.

3. Run scripts/generate_cv_selection.py --job <job.json> --output <cv_selection.json>. Use --limits <limits.json> for reviewed CV limits, and optionally --base <base/cv_selection.json> --mode adaptive|strict for a derived application. Review its deterministic suggestion and adjust only with facts supported by the profile.

4. Create one application_content.json with schema_version: 2, job, cover_letter, and plain_text as in examples/application_content.json. Keep claims grounded in selected profile records. The legacy cover_letter.json and indeed_text.json adapters are transitional; use scripts/convert_application_content.py to migrate them. The cover-letter template renders job.company as its own first recipient line. Therefore cover_letter.recipient must name only the contact or team, and cover_letter.company_address must contain only address/location lines, never the company name again.

5. Generate source documents:

   python3 scripts/generate_cv.py --selection <cv_selection.json> --output <CV.tex>
   python3 scripts/generate_cover_letter.py --content <application_content.json> --output <cover_letter.tex> --profile-dir <profile-directory>
   python3 scripts/generate_plain_text.py --content <application_content.json> --output <indeed_text.txt>
   

6. If pdflatex is available and PDFs are requested, run the cross-platform renderer:

   python3 scripts/render_application.py <application-dir>
   

   Rendering runs validate_application.py by default; Poppler's pdfinfo and pdftotext are required. The visible --skip-validation switch is for debugging only. Use validate_application.py <application-dir> --strict-fill to make an underfilled CV fail. For an external profile, pass the same --profile-dir to the cover-letter generator and renderer (or configure JOB_APPLICATION_PROFILE_DIR). The generator refuses an external letter that would silently use the bundled fictional profile, and the validator checks the letterhead against the selected profile. On Windows, py -3 can be used instead of python3. Linux and macOS users may alternatively use the scripts/render_application.sh wrapper.

7. Return the generated paths and validation result. Deliver files externally only after destination-specific approval.

   • If the user explicitly asks to receive the documents in the current chat and the channel supports attachments, send the validated CV.pdf, cover_letter.pdf, and (when useful) indeed_text.txt as native downloadable document attachments in that conversation.
   • Do not present local filesystem paths or file:// links as downloadable chat links. They are references for the local host only and are not usable by most chat clients.
   • Never attach source JSON, profile data, signature assets, credentials, or intermediate files. Attach only the final requested deliverables after validation has passed.
   • Treat every chat attachment as destination-specific external delivery. A request such as “send the documents here” authorizes only that current conversation, never an application submission or delivery to a recruiter.

Installation requirements and platform-specific commands are documented in README.md. Generated sample PDFs using only fictional data are in examples/generated/.

Content policy

• Default to a one-page CV.
• Default CV limits are five skills, three experience entries, two education entries, one project, three languages, and two achievement sentences per experience. They may be intentionally changed only in the reviewed selection limits; the generator enforces them.
• Prefer evidence relevant to the posting over generic claims.
• Keep each experience entry to the reviewed achievement_sentences_per_experience limit (two by default) and make every sentence concise.
• Preserve optional project attribution fields (organization, context, and context_label) so generated CVs can distinguish employer projects from work performed through the candidate's own company.
• Include the profile's available address or location by default, including a street address when supplied. Omit date of birth and a signature image by default. Include a signature image only when the user requests it through cover_letter.signature_image. Include a profile photograph only when the user requests one or the supplied profile explicitly declares profile_picture; the bundled photo is fictional demonstration material.
• Match the job's language unless the user requests another language.

JSON contains content, Python performs deterministic transformation, and LaTeX handles layout. Do not hand-edit generated LaTeX when the source JSON can be corrected instead.