What problem does it solve? Observability-driven development runs lose their findings between sessions because reports, benchmarks, and maintainer decisions have no committed home, forcing every new run to start blind. This Skill defines the contract for a git-committed memory under .odd/ so reports, benchmarks, custom stacks, and rulings are versioned, reviewable, and recallable. ## Core Features & Use Cases - Memory contract for five artifact kinds: observation reports, instrumentation reports, maintainer-ruling ledgers, k6 benchmarks, and custom observability stacks, each with its own reference file defining paths, frontmatter, recall matching, and commit discipline. - Scripted recall and ledger writes: scripts/odd_recall.py filters stored frontmatter by service, stack, and environment without loading prose into context, and scripts/odd_ledger.py validates and appends finding decisions and tree-entry classifications. - Append-only evidence with living-source exceptions: reports and ledgers are never rewritten, while benchmarks and custom stacks update through reviewed diffs, with work-branch and lone-commit rules enforced. - Use Case: After an observation run finds a latency regression, persist the report to .odd/observe-run-reports/, record a wontfix decision on one finding via the ledger script, and recall the baseline before the next verification run. ## Quick Start Ask the agent to recall the previous observation reports for a service and stack using the odd-memory recall script before starting a new observability run.
