irl-dan
2de0b5e7d7
Narrow intake a founder can run in under a minute. Three prompts, one WebFetch batch across a fixed set of well-known paths, one screenshot-ready markdown report written to ~/prose-<company>/agent-readiness.md. Scores four dimensions (discoverability, parseability, structure, api-readiness) 0-100 and surfaces exactly three concrete, file-path-level fixes. No JavaScript execution, no fetches outside the target origin, no writes outside the per-company workspace. Built for the "claude \"prose run co/programs/agent-readiness\"" first experience: all probes are static well-known paths, so it never fails on a live JS-wall or a missing API. Paired eval protects the demo: every finding must ground in a Raw Probe Results row, every top_fix must name an exact file path, and any JavaScript execution or cross-origin fetch fails the eval. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3.1 KiB
co — Company as Prose
Reusable starter contracts for organizing an operating company as an OpenProse-native repository.
co sits next to std under packages/, not inside it. std is the
low-level standard library: roles, controls, delivery adapters, memory, evals,
and ops primitives. co is an opinionated starter kit for a specific domain:
a company whose operating system is made of Prose programs.
Reference programs in co with the co/ shorthand (analogous to std/):
use "co/programs/company-repo-checker"
Which expands to github.com/openprose/prose/packages/co/programs/company-repo-checker.
Package Shape
packages/co/
README.md
programs/
agent-readiness.md
company-repo-checker.md
evals/
agent-readiness.eval.md
company-repo-checker.eval.md
Programs
agent-readiness— narrow intake a founder can run in under a minute: scores how accessible their site is to AI agents (well-known paths, plain-HTML parseability, structured metadata) and writes a screenshot-ready markdown report to~/prose-<company>/agent-readiness.md. Safe first-run demo: three prompts, one WebFetch batch, one report.company-repo-checker— static gate that verifies a company-as-prose repository still matches the shared layout before expensive runtime evals or fleet monitors do more work.
Future additions should help a new company get started without copying OpenProse, Inc.'s private business logic:
- starter repository architecture
- company system map
- customer package checker
- eval ladder
- fixture and run-replay conventions
- onboarding workflow for the first operating responsibilities
std vs co — the split
- std — use-case-agnostic primitives. Inspector, contract-grader, retry, fan-out, worker-critic, human-gate. Things that make prose programs work.
- co — company-operations-shaped patterns. Starter repo checkers, scheduled intake, windowed analytics, GTM pipelines, fleet monitors. Things that make prose programs produce business value.
Running Programs
prose run is an agent-session command. It is not assumed to be a shell binary.
If a host provides a native Prose CLI, use it. Otherwise wrap the command in an
agent runner that has the OpenProse skill loaded.
Claude Code:
claude -p "prose run co/programs/company-repo-checker --repo_path customers/prose-openprose"
Codex:
codex exec -C <workspace-root> "prose run co/programs/company-repo-checker --repo_path customers/prose-openprose"
The shell executable is claude or codex. The prose run ... string is the
instruction the agent session interprets as the OpenProse VM.
Design Notes
- Keep this package generic. Do not include OpenProse, Inc. leads, accounts, GTM logic, release logic, or private operating assumptions.
- Prefer composable programs with inline starter components until a component earns a stable public API.
- Put universal primitives in
std/; put company-operating-system patterns here. - Keep generated runtime state out of this package. Useful lessons can become docs, fixtures, or evals.