value-pricing
GitHub stars Follow C-Kapsalis on GitHub
An open-source agent skill and plugin for a rigorous value-based pricing study of any product or service. Give it a website (and optionally an Obsidian vault, a docs folder, decks or past artifacts); it builds the capability vector with evidence, prices it from three sides, estimates each feature's added value, convenes a council of persona agents, runs a pre-registered simulated-buyer panel, interviews you about your cost structure, and scores packaging proposals on contribution after people time.
Every price it produces is a hypothesis to validate with real buyers. It says so, and phase 7 designs that validation.
What's inside
| Path | What |
|---|---|
skills/value-pricing/SKILL.md | the orchestrator: ground rules, workspace, the seven phases with completion criteria |
skills/value-pricing/references/ | one file per phase, plus lessons.md, literature.md, output-contract.md |
skills/value-pricing/templates/ | study.yaml, offering.yaml, taxonomy.yaml, value_drivers.yaml, costs.yaml, proposals.yaml, plans.csv, vendors.csv, preregistration.md, DECISIONS.md |
skills/value-pricing/scripts/ | the analysis code (Python 3.11+), each script with --help |
agents/ | the council: buyer-persona, pricing-strategist, cfo-unit-economics, behavioral-economist, competitive-analyst, gtm-sales-lead, devils-advocate |
commands/value-pricing.md | /value-pricing <url> [paths…] |
The method
| Phase | Script(s) | Output |
|---|---|---|
| 1 Intake | crawl_site.py, ingest_corpus.py | offering.yaml with evidence, taxonomy.yaml, segments and archetypes ruled by the owner |
| 2 Market | crawl_site.py --vendors, code_features.py, llm_batch.py, kappa.py, hedonic.py, value_metric.py | implied price per segment, market premium per feature, what rivals meter on |
| 3 Value | value_mc.py | EVC per segment, per-feature added value (share valued, Shapley share, leave-one-out), Sobol |
| 4a Council | the agents | objections, fences, behavioural parameter ranges, verdicts per structure |
| 4b Panel | conjoint_design.py, panel_build.py, llm_batch.py, panel_analyze.py, conjoint_fit.py, calibration.py | pre-registered conjoint WTP, Van Westendorp, Gabor-Granger, anchor effect, level-bias ratio |
| 5 Cost | cost_model.py | floor per segment, people time, fixed-cost carry, support minutes per price |
| 6 Synthesis | synthesize.py, optimize_packages.py | price corridor, feature-value classes, proposals tuned on contribution under every level lens |
| 7 Validation | — | real-buyer survey with decision rules fixed in advance; the owner's rulings |
It grew out of real pricing work; references/lessons.md records what that work taught.
Install
Claude Code plugin
Add the GitHub marketplace and install the plugin:
/plugin marketplace add C-Kapsalis/value-pricing
/plugin install value-pricing@value-pricing
You can also download value-pricing-plugin.zip from the
latest GitHub release
and install it as a local plugin.
Codex or another Agent Skills client
Install the skill from the public repository with:
npx skills add C-Kapsalis/value-pricing
Or download value-pricing-skill.zip from the
latest GitHub release
and place the value-pricing/ folder in that client's skills directory.
Codex users can also add this repository marketplace with
codex plugin marketplace add C-Kapsalis/value-pricing and install the
Value Pricing plugin from the marketplace.
ChatGPT and Codex public directory review uses the
value-pricing-openai.zip archive; a public listing is available only after
vendor review and publication.
Python environment (once per machine):
uv venv .venv-pricing
uv pip install --python .venv-pricing/bin/python -r <path>/skills/value-pricing/scripts/requirements.txt
# JavaScript-heavy sites: uv pip install playwright && .venv-pricing/bin/playwright install chromium
The panel and feature coding call claude -p on your Claude plan by default (no API key spend);
--runner api uses the Anthropic API when you choose to.
Use
/value-pricing https://yourproduct.com ~/Obsidian/YourVault ./sales-decks
or just ask: "price my product", "what should we charge", "design our tiers", "what is each feature worth to customers". The skill asks before every expensive step and every judgement call.
Tested
scripts/selftest.py runs every analysis script on a synthetic study with planted ground truth (feature
premia, a meter elasticity, simulated buyers with known willingness to pay) and prints PASS/FAIL per
check. Run it once after installing: .venv-pricing/bin/python <path>/scripts/selftest.py.
At packaging time the hedonic model recovered all three planted premia inside its 90% intervals, the meter table recovered the planted elasticity (0.73 vs 0.75), and the HB conjoint recovered the planted willingness to pay per level within a few euros. The crawler and corpus ingest were run on real material. A full live study (hundreds of LLM calls for feature coding and the panel) was not run as part of packaging.
Distribution
See PUBLISHING.md for archive contents, release instructions, marketplace setup and vendor-directory submission steps.