import-claude-context: prompts/session.md is missing but SKILL.md step 5 requires it

Start at skills/import-claude-context/SKILL.md step 5 and compare skills/import-claude-context/prompts/project.md and entities.md with finalize.py:415. Add the missing prompts/session.md with the authoritative session-summary schema, then verify a fresh import can complete step 5 and produce summaries consumed by the downstream prompts.

What happened?
SKILL.md step 5 of skills/import-claude-context instructs: 'one Agent-tool subagent per session with model: haiku, prompt = prompts/session.md with the slug/uuid/chunk paths filled in'. That file does not exist in the shipped skill.

skills/import-claude-context/prompts/ contains only entities.md and project.md. The workspace copy at workspace/.claude-sutando/skills/import-claude-context is a symlink to the engine tree, so there is no second copy that supplies it.

Impact: the per-session summarisation step — the bulk of the import — has no prompt template. finalize.py only reads 'title' from summaries//.json directly (finalize.py:415); the rest of the session-summary schema is consumed by prompts/project.md and prompts/entities.md, so an agent has to reverse-engineer the schema from those two downstream prompts to proceed. Different agents will infer different schemas, and a field project.md/entities.md expect but a worker omits degrades the roll-up and entity passes silently.

Repro: fresh install, run /import-claude-context (or the onboarding Import button), reach step 5, then: ls skills/import-claude-context/prompts/ -> entities.md, project.md.

Workaround used on this host: briefed the per-session workers with an inline schema derived from project.md + entities.md + finalize.py rather than writing session.md into the engine tree (which is replaced on app update).

Suggested fix: ship prompts/session.md, mirroring the structure of project.md/entities.md, with the session-summary schema stated once and authoritatively.

Evidence: engine v as installed 2026-09-10; skills/import-claude-context/SKILL.md step 5; skills/import-claude-context/prompts/ listing; finalize.py:415.

Secondary, lower confidence and probably environmental rather than a code defect: secret_scanner reports 'mode: DEGRADED — detect-secrets missing; repo-local whole-line rules only' on every index/extract invocation. detect-secrets is not installed and this host's python3 is Homebrew-managed (PEP 668 externally-managed), so pip install fails without --break-system-packages. Extract still redacted 8 items. If full-strength redaction is expected on a normal install, the dependency may need a bundled/isolated install path rather than relying on the system Python.

Steps to reproduce

1. (from reporter — to confirm)

Expected behavior
(to confirm)

Severity (reporter): medium

_{🤖 Filed automatically from in-app user feedback — bug / severity medium. Edit freely; this is a starting point, not a fixed record.}