Structured cognitive discipline
Most AI-assisted work loses its reasoning the moment the diff lands. StrayMark fixes that with three small artifact types — Charters, AILOGs, AIDECs — that sit next to your code and document the intent, the action, and the trade-off behind each change.
Why this matters
When an agent ships a feature, you usually get the result and lose everything that led there: the alternatives considered, the scope explicitly excluded, the risks discovered mid-flight. Six months later, the code is unrecoverable archaeology. Cognitive discipline is cheap when you have a template; it's everything when you don't.
The cost of not having it shows up later as: undocumented decisions that nobody can defend in a review, agents drifting silently from the spec, and compliance auditors asking for evidence that doesn't exist.
What you get
- Charters declare a bounded unit of work before you start: scope (in / out), files declared up-front, risks (
R1...Rn), verification commands. The Charter cannot close if reality drifted from the declaration and the drift wasn't reconciled in the same PR. - AILOGs log the actual execution: what got done, why, what was discovered in flight, what was punted. One per commit, sequence-numbered per day, with a
risk_leveland aconfidencefield. - AIDECs capture standalone decisions: alternatives considered, trade-offs, the option chosen and the reason. The decision survives even if the original Charter is closed.
All three are plain Markdown with YAML frontmatter. They live in .straymark/ and ship with templates. The CLI scaffolds them; agents invoke them via skills.