chameleon

name	chameleon
description	Tailor the master resume for a specific job posting, persisting the job analysis locally before updating the CV. Use when the user provides a job URL or pasted job description and wants a tailored CV generated.
disable-model-invocation	true

/chameleon

Tailor a master CV YAML to a specific job posting, saving the structured analysis locally first.

Usage

/chameleon <job-url-or-paste> [--cv <master-cv-name>]

Examples:

• /chameleon https://example.com/jobs/123
• /chameleon --cv john_doe "Senior Engineer at Acme Corp..."

Workflow

Follow these steps exactly in order:

Step 1 — Obtain the job description

If the argument is a URL, fetch its content using WebFetch. If it is pasted text, use it directly. Store the raw JD text for the next step.

Step 2 — Analyze the job description

Spawn the analyze-job-posting agent. Pass it the full raw JD text. Wait for the structured analysis output before proceeding.

The agent must return a JSON object with these fields:

• company_name
• role_title
• seniority
• required_skills
• preferred_skills
• responsibilities
• ats_keywords
• positioning_signals
• summary_angle

Step 3 — Save the analysis artifact

Persist the agent's JSON output locally under output/job_analyses/.

The saved filename must follow:

<analysis_id>__<company_slug>__<role_slug>.json

Do not write the artifact under templates/.

Step 4 — Resolve the master CV

Scan the templates/ directory for files matching the pattern *_cv.yaml.

• If --cv <name> was passed, use templates/<name>_cv.yaml. If it does not exist, report the error and stop.
• In all other cases — whether one file exists or many — list all found *_cv.yaml files and ask the user which one to use before continuing. Never auto-select silently.

Step 5 — Tailor the CV

Spawn the update-cv-with-job-posting agent. Pass it:

1. The resolved saved analysis JSON
2. The resolved master CV file path

Wait for the agent to save the tailored YAML to templates/. The agent does not render — it only writes the file.

Summary Quality Bar

The tailored summary must read like a resume summary, not a recruiter recommendation.

• Lead with the candidate's actual experience and strengths, not with a long stack list.
• Never use recruiter-facing evaluation language such as strong fit, ideal candidate, should be shortlisted, for this role, or this candidate.
• Write in third-person-neutral resume style focused on the person's experience and skills. Do not use I, and avoid pronoun-led phrasing when a direct skills-first sentence is cleaner.
• Show relevant expertise, but connect it to customer, business, or real-world impact whenever the master CV supports that framing.
• If the master CV and JD support it, surface genuine motivation for the problem space, product, or mission in concrete language rather than generic enthusiasm.
• Keep the first paragraph focused on technical fit and strongest role-relevant strengths.
• Use the second paragraph for motivation and broader impact, and include at least one concrete metric grounded in the master CV when one is available.
• Keep the summary to at most 2 paragraphs.
• Keep the broader stack coverage in the skills section instead of forcing it into the summary.

Layout Defaults

Tailored resumes should preserve the master CV's design block unchanged. In this repo, the compact classic-theme defaults include design.typography.line_spacing: 0.8em, design.typography.font_size.body/headline/connections: 9.5pt, and design.sections.space_between_regular_entries: 0.2cm.

Step 6 — Render the tailored CV

This step is mandatory and must always run after Step 5 completes, even if the agent did not report errors.

Use the Makefile render target directly:

make render FILE=templates/<username>_<company>_<role>_cv.yaml

Where <username> comes from the selected master CV filename with the trailing _cv removed, and <company> and <role> come from the analysis output. Lowercase all parts, replace spaces with underscores, and remove special characters.

Step 7 — Report to the user

If the render succeeds, report:

• analysis ID
• saved JSON path
• company name
• role title
• tailored YAML path
• generated PDF path

If the render fails, show the full error output so the user can act on it. Do not silently swallow errors.

Error Handling

• If the URL cannot be fetched, report the error and ask the user to paste the JD text directly.
• If rendercv is not installed, tell the user to run make install-tools and retry.
• If saving the analysis artifact fails, report the error clearly.
• Never silently skip a step. If any step fails, stop and report what went wrong.