Summary
Point combines a semantic language surface with compiler APIs that expose stable context to coding agents.
Semantic blocks are the source
In Point, semantic declarations are not comments or annotations on top of another language. They are the source language.
record Launch Signals
has bundle id: Bool
submitted for review: Bool
has passing tests: Bool
rule launch readiness
input signals: Launch Signals
output score: Int
score starts at 0
add 30 when signals.has bundle id
add 40 when signals.submitted for review
add 30 when signals.has passing tests
return scoreThat source is compact enough for humans to read, but structured enough for a compiler and agent to index.
Stable refs beat line numbers
Agents often fail because they patch by nearby text, line numbers, or generated names that shift under formatting. Point gives public declarations stable semantic refs:
point://semantic/Math/record.User
point://semantic/Math/rule.launch readiness
point://semantic/Math/label.score statusThese refs survive formatting and small edits better than raw source positions.
Repair loops are first-class
The CLI emits structured diagnostics through check-json and ordered repair steps through repair-plan. A coding agent can read ref, expected, actual, repair, and relatedRefs, patch the semantic source, and run the checker again.
That is different from asking an agent to paste a whole file into a prompt and hope it notices the right part.
Explicit effect boundaries
Point separates pure blocks from effectful blocks. external declarations mark interop boundaries, action blocks carry effect metadata, and policy blocks express pure guards.
This gives review tools and agents a clearer map of when code reaches the filesystem, network, environment, time, randomness, or external packages.