Skip to content

Skill: primitives, semantics, limits; reference for fields and codes - #17

Merged
bencode merged 1 commit into
mainfrom
skill-coverage
Oct 5, 2026
Merged

bencode merged 1 commit into
mainfrom
skill-coverage

Conversation

@bencode

@bencode bencode commented Oct 5, 2026

Copy link
Copy Markdown
Contributor
  • SKILL.md: access, the seven endpoints (one line each), semantics an agent cannot infer, and what skym cannot tell (metric history, logs, processes, who did something, acting). No question-to-endpoint recipes: the agent composes the primitives itself.
  • reference.md (new): parameters and response shapes, fields per view, event types, exception groups, the 14 incident codes with thresholds and what to check next. Read on demand.

Verified by a fresh agent with only these two files and a reader token, against the live API: 15 questions, including three skym cannot answer (declined correctly). Its findings were fixed and rechecked.

SKILL.md lists each endpoint once, the semantics an agent cannot infer
(verdicts already judged, UTC, open_for against since, muting, hygiene,
business exceptions, the lower bound of other memory) and what skym cannot
answer, so an agent says so instead of guessing. reference.md holds the
facts to look up: parameters, response shapes, fields, event types,
exception groups and the incident codes with what to check next. Checked
by an agent given only these files against the live API.
@bencode
bencode merged commit f44880e into main Oct 5, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant