session-historianlisted

# /session-historian — what did we already try? Every significant investigation leaves traces in `memory/`, the self-review correction log, `memory/compounded/`, and the session transcripts themselves. This skill surfaces those traces so you don't re-walk dead ends. Inspired by Every's `ce-session-historian` — stripped down for our single-harness (Claude Code only) setup. ## When to invoke - Starting a task that sounds familiar ("didn't we already look at this?") - About to propose an approach that might have been tried before - Debugging a repeat correction (the rule is there; why isn't it landing?) - Before `/compound`-ing a correction, to check if it's a fresh shape or a recurrence of a known one ## Do NOT invoke for - General project orientation (that's what CLAUDE.md + MEMORY.md are for) - Fresh topics with no likely history — the skill returns nothing useful and costs attention ## The flow ### Step 1 — Normalize the query Accept `$ARGUMENTS[0]` as a topic/keyword. Expand to 3–5 related search terms. Example: - Input: "AE101 delivery" - Expanded: AE101, delivery architecture, knowledge home, cohort, Teacher Claude ### Step 2 — Search in priority order Parallel searches (single message, multiple Grep calls): 1. **`memory/compounded/`** — schema-validated corrections/patterns. Highest signal. 2. **`memory/project_*.md`** — decisions and investigations. 3. **`memory/self-review-protocol.md`** — correction log. 4. **`memory/feedback_*.md`** and legacy feedback fi