Selvedge vs. Origin
Origin (opsworks-co/origin-cli) captures automatically — agent lifecycle hooks record prompt receipts live, per turn, anchored to commits in git notes. Selvedge captures the agent’s stated reasoning through MCP. The trade is coverage vs. content: receipts everywhere, or testimony where it matters.
| Selvedge | Origin | |
|---|---|---|
| Reasoning source | Captured live, by the agent in the same context that produced the change | Prompt receipts, captured live per turn — prompt, diff, tokens; no stated rationale |
| Granularity | Entity — DB column, table, env var, dep, API route, function | Line |
| Mechanism | MCP server — the agent calls it as work happens | Agent lifecycle hooks + global git post-commit hook |
| Grouping | Changesets — named feature/task slugs across many entities | Branch-level trail view; no entity-spanning changesets |
| Prior attempts | Yes — prior_attempts surfaces tried-and-rejected paths | None — rework flags reverted AI code post-hoc, without rationale |
| Storage | SQLite, zero deps | Git notes (refs/notes/origin) + a sessions branch; optional cloud sync |
Where Selvedge is different
Section titled “Where Selvedge is different”- Receipts vs. testimony. Origin records what the prompt and the diff were — automatically, across a dozen-plus agents. Selvedge records what the agent said it was doing and why, from the context that had the reasoning. A receipt can tell you a change followed a prompt; it can’t tell you which alternatives were weighed and killed.
- Rejected paths. Origin’s
reworkcommand detects AI code that was later reverted or heavily modified — a post-hoc diff heuristic yielding a rework-rate report, with no rationale attached and no queryable record of why the approach died. Selvedge’sprior_attemptsreturns the tried → reverted → re-opened trail with the stated reason at each step. - Entity-level granularity. Selvedge’s unit is the column / env var / route / dependency / function, with prefix queries — not lines that drift as the file evolves.
- Read path for the agent.
prior_attempts,blame,diff, andstale_decisionsare callable by the agent mid-task, and the SessionStart digest pushes standing verdicts into new sessions. Origin’s queries (origin why,origin blame, session search) are developer-facing; nothing pushes a prior verdict back into the agent’s context.
When Origin might fit you better
Section titled “When Origin might fit you better”If you want automatic, zero-cooperation capture across a dozen-plus agents — prompts, diffs, and token costs riding along in git notes with optional cloud sync — Origin’s coverage and near-zero adoption friction beat an MCP-cooperation model. Reach for Selvedge when what you need back is the reasoning and the rejected paths, not the receipts.