Runs a complete therapeutic drug-development program for any disease — end to end, scientist in the loop.
A specialist agent profile and its companion skills for Claude Science. Runs an end-to-end, human-in-the-loop therapeutic drug program for any disease: unmet need -> targets/modality -> AI/ML design -> scientific/regulatory/commercial strategy -> deliverables.
agent/
profile.json # picker metadata (name, description, access)
system_prompt.md # the agent's identity / opening system prompt
agent_manifest.json # (legacy) manifest from the original release
skills/ # 9 skills, each with SKILL.md, kernel.py, and any assets
install.py # one-shot, idempotent installer (recommended)
deploy_agent.py # (legacy) original deploy script from the first release
- agentic-campaign-manuscript
- antigen-epitope-pipeline
- drug-program-orchestration
- drug-repurposing-scout — screen approved drugs for repurposing signals in a disease (mechanism/target overlap, published reports, existing trials) and explain expanded-access / compassionate-use as neutral process information. The catalog's most safety-sensitive skill:
H-SAFETY-defining — never a treatment recommendation, dose, or claim any drug works; approved drugs only; individual decisions route to the patient's physician + FDA + IRB. - omics-target-mining
- patient-centered-market-and-survey
- scientific-poster
- synthetic-peer-review
- systematic-review-orchestration
Please install the Therapeutic Program Architect from this GitHub repo: https://github.com/ruthannepai-tech/therapeutic-program-architect Clone or download it, then run
install.pyfrom the repo root in the repl tool withexec(open("install.py").read()). It will create and publish the skills and create the DRUG_PROGRAM_ARCHITECT agent profile with full access. Then create its environment and offer to switch me to it.
When it finishes, Therapeutic Program Architect appears in your agent picker.
From the repo root, in the Claude Science repl tool:
exec(open("install.py").read())install.py publishes every skill under skills/ (with all their asset files)
and creates the DRUG_PROGRAM_ARCHITECT agent profile with full catalog + connector access. It
is idempotent — safe to re-run; it updates in place. It finishes by printing
the optional environment setup and the switch command.
The skill files and agent/ files are plain text and fully define everything.
(deploy_agent.py from the original release is preserved for reference; the
install.py route above is the maintained one.)
Some skills here are shared with other agents I've published, so they're bundled in each repo that needs them. If you improve one, the same change may be worth applying wherever it appears:
agentic-campaign-manuscriptantigen-epitope-pipelineomics-target-miningsynthetic-peer-reviewsystematic-review-orchestration
MIT — see LICENSE.