Pre-edit memory
Agents read relevant decisions, risks, constraints, and anti-patterns before non-trivial edits.
Git-native intent memory
Stop coding agents from repeating old engineering mistakes.
Mainline gives AI coding agents the histórical why before they edit the current what: abandoned approaches, superseded decisions, reviewer constraints, risks, and in-flight intent.
Install
curl -fsSL https://raw.githubusercontent.com/mainline-org/mainline/main/install.sh | bash $ mainline context --current --json
{
"anti_patterns": [
{
"what": "do not remove legacy /oauth",
"why": "callbacks still require session state",
"severity": "high"
}
],
"risks": ["billing migration still has duplicate-event risk"]
}Intent sealed
Replace session auth with JWT
- Decision
- JWT for mobile stateless auth
- Rejected
- Remove OAuth middleware
- Pin
- refs/notes/mainline/intents
0
forbidden-list violations in intent-first eval runs
9
violations from code-first agents across the same scenarios
Git
stores repo memory in refs and notes, not one vendor workspace
The failure mode
Reasonable code. Wrong histórical reason.
A coding agent can grep the repo, inspect a TODO, and finish the half-built Redis queue. That may still be the wrong move if the team already abandoned Redis because replication lag caused duplicate billing events.
Code search tells agents what exists now. Mainline records the decision memory that explains what should not be repeated.
Review intent
Reviewers compare the implementation against the stated why instead of inferring intent backward from a diff.
Intent governance
Teams can see coverage gaps, low-quality sealed intents, risky changes, and overlapping in-flight work.
Long-term memory
Future maintainers know which approaches were tried, abandoned, superseded, accepted, or pinned to commits.
Agent loop
Read prior intent. Make the change. Record new intent.
mainline context --current --jsonRetrieve repo memory before edits.mainline start "Add JWT auth"Claim a real unit of work.mainline append "Implemented middleware"Record decisions, pivots, and risks.mainline sealPreserve the durable decision record.
For solo builders
Preserve continuity between agents, branches, and future sessions. Your next agent should know why the last one abandoned an approach.
For teams
Make repo truth shared before review. See proposed work, file-level constraints, open risks, and intent coverage without Slack archaeology.
Public alpha
Coding agents should inherit engineering memory.
Mainline is usable today for early teams that want Git-native repo memory before agent-generated changes reach review.