story-synclisted
Install: claude install-skill vitdo82/story-drift
Argument: a Jira key, `--all` for your working set, or `--epic PROJ-100` for one epic's whole tree.
## Why
An AI agent that has been working on a story for three days remembers its **interpretation** of the requirements, not the requirements. When the PO edits the acceptance criteria on day two, the agent keeps building the version it read on day one — and no amount of "re-read the ticket" fixes this, because a re-read has nothing to compare against.
A snapshot with per-field hashes has something to compare against.
## Scope — strictly three rings, never the whole project
- **My stories** — every `.story/*/context.json`. Full diff: fields *and* comments.
- **Parent epic of each of my stories** — description, acceptance criteria, priority.
- **Sibling stories in those epics** — every child of the epic, whether or not it is linked to mine. Fields plus the last comment author only.
The sibling ring is the one people forget: a sibling changing its acceptance criteria can change your contract, so it is signal even though you never opened that ticket. Nothing outside these three rings is ever fetched.
## Steps
1. **Re-fetch** only the three rings. Hash the same fields with the same normalisation (strip whitespace and markdown).
2. **Diff** against the snapshot. Do not re-read the whole story into the conversation — report only differences.
3. **Impact.** For a change to *my* story, run `git diff --stat` / `git diff --name-only` on the current branch and say which files it