From orientation
Use when judging or planning around code not written this session: repo state, next steps, bug/dead-code calls, cleanup, or adding near existing code. Run orientation history before git/diff/README recon; after compaction, treat pre-compact work as unfamiliar.
How this skill is triggered — by the user, by Claude, or both
Slash command
/orientation:get-orientedThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
When you land in code you did not write this session, you do not know its
When you land in code you did not write this session, you do not know its story yet. The trap is reading the code, README, diffs, and tests and then reconstructing a plausible story. That reconstruction is often wrong. A tidy README and a clean compile can make abandoned work look finished, or make active work look stale.
This skill pulls the real orientation history from normalized agent work events: the human goal behind the code, whether it shipped, how recently it was touched, and which files moved with it.
Keep this skill's YAML frontmatter to name and description only. Do not put
tool allowlists, arrays, nested objects, or host-specific metadata there. Some
hosts accept richer skill metadata and others reject it; this skill needs to
load everywhere. Treat the commands below as usage guidance, not frontmatter
schema.
Expected tool families when the host supports scoped permissions:
git rev-parseorientationnodeThe trigger is mechanical, not emotional. Ask one thing: did I edit this code myself in this un-compacted session, and do I still recall its history? If no to either, and you are about to answer a question about it or change it, run orientation first.
Always run it before ordinary git/readme/diff recon when:
After a compaction, "I wrote this earlier" is not a valid skip. Compaction replaces the earlier turns with a lossy summary. Treat pre-compact work like someone else's history and ask orientation for the actual goals and recency.
git rev-parse --show-toplevel
orientation provenance "<root>" "<file>"
orientation related "<root>" "<file>"
Use provenance when judging, deleting, or understanding existing code. Use
related before adding near existing code.
If orientation is not on PATH, use the installed engine fallback. Prefer
the Eigen-local runtime when present, then the legacy compatibility path:
node ~/.eigen/orientation/consume.js --provenance "<root>" "<file>"
node ~/.eigen/orientation/consume.js --related "<root>" "<file>"
node ~/.claude/action-graph/consume.js --provenance "<root>" "<file>"
node ~/.claude/action-graph/consume.js --related "<root>" "<file>"
<file> may be a relative path, basename, or absolute path.
The verdict line is recency-first. Uncommitted work is not abandoned by default; in-flight work is usually uncommitted.
ACTIVELY IN FLIGHT: someone is delivering this now. Continue it or ask;
do not scrap it as stale.DELIBERATE WORK: committed, intentional work. Do not delete as cruft
unless callers, exports, and tests prove the specific code is dead.POSSIBLY STALE: uncommitted and untouched for a long time. Verify with the
user or real signals before treating it as dropped.No recorded work: silence is not permission. Judge from callers, exports,
tests, and behavior.Coupled neighbors: files usually edited alongside this one. Check them
before treating the target as isolated.Read the goals that already built in this area. Extend or reconcile with those decisions instead of duplicating or contradicting them.
Read sibling files that orientation reports. Grep those files and the target for the thing you are about to add before writing it.
Say what the history showed and how it changed your read. If the orientation verdict and the code seem to conflict, surface that conflict with the goal cited instead of acting on your reconstruction alone.
npx claudepluginhub agent-sh/agentsys --plugin orientationCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.