Support Copilot CLI multi-storage recall and provider-based fallback

[jshessen]

Summary

This PR resolves Copilot CLI 1.0.34+ storage compatibility by introducing provider-based multi-storage recall instead of assuming a single SQLite layout.

Closes #3

Storage model coverage by provider

Provider	Storage location(s)	Storage format	Commands supported
cli (Copilot CLI)	~/.copilot/session-store.db	SQLite (legacy/current where present)	list, search, show, files, checkpoints, health, schema-check
cli fallback	~/.copilot/session-state/*/events.jsonl	JSONL events	list, search, show (schema-check reports compatibility mode)
vscode	~/.config/Code/User/workspaceStorage/**/chatSessions/*.jsonl (+ flatpak/snap variants)	JSONL logs	list, search, show, files
jetbrains	~/.config/github-copilot/chat-sessions/* + related chat files	file-backed JSON/session logs	list, search, show, files
neovim	~/.config/github-copilot/** and ~/.local/share/nvim/** chat JSON/JSONL	file-backed JSON/JSONL	list, search, show, files

Environment overrides supported

• SESSION_RECALL_DB
• SESSION_RECALL_CLI_STATE_ROOT
• SESSION_RECALL_VSCODE_STORAGE
• SESSION_RECALL_JETBRAINS_ROOT
• SESSION_RECALL_NEOVIM_ROOT

Runtime behavior and fallback semantics

Provider discovery

At runtime, providers are discovered and filtered to those actually available on the machine:

• CLI provider is available if either SQLite DB exists or session-state event files exist.
• File-backed providers are available when their roots exist.

Command behavior in multi-storage mode

• list, search, show, files, checkpoints now route through active providers.
• schema-check is provider-aware:
  • Performs strict schema checks when CLI SQLite is present.
  • Returns compatibility detail (session-state-or-sqlite) when fallback event storage is active.
• health is provider-aware:
  • Runs SQLite health dimensions when SQLite is available.
  • Falls back to provider compatibility dimensions when SQLite is not available.

UX and recall quality improvements

• Added session-recall repos to summarize discovered repositories/workspaces across providers.
• Improved repository attribution for CLI session-state parsing.
• Improved list/repos alignment and sparse-scope behavior so recall output better matches discovered session reality.

What changed

• Added provider architecture:
  • src/session_recall/providers/ (base.py, discovery.py, copilot_cli.py, file_backends.py, common.py)
• Added repo discovery command:
  • src/session_recall/commands/repos.py
• Updated command paths for provider fallback and multi-storage context:
  • list, search, show, files, checkpoints, health, schema-check
• Improved repo detection/output formatting:
  • src/session_recall/util/detect_repo.py
  • src/session_recall/util/format_output.py
• Added/updated tests:
  • test_provider_backends.py
  • test_health_schema_multistorage.py
  • test_repo_scope_fallback.py
  • test_repos_command.py
  • updates in test_list_sessions.py

How this differs from PR #4

PR #4 primarily introduces a DB-layer adapter approach (db/jsonl_store.py + connect/schema integration).

This PR addresses the same issue via provider routing and command-layer fallback, including:

• dynamic provider discovery,
• explicit session-state support for CLI,
• cross-surface file-backed providers (VS Code / JetBrains / Neovim),
• and provider-aware health/schema semantics.

Both aim at Issue #3 but with different architecture and operational surface area.

Validation

• pytest src/session_recall/tests/ -q → 105 passed
• ruff check src/ → all checks passed
• Focused suite:
  • pytest -q src/session_recall/tests/test_provider_backends.py src/session_recall/tests/test_repo_scope_fallback.py src/session_recall/tests/test_repos_command.py src/session_recall/tests/test_list_sessions.py src/session_recall/tests/test_health_schema_multistorage.py → 18 passed

CONTRIBUTING checklist

• Tests pass: pytest src/session_recall/tests/ -q
• Lint passes: ruff check src/
• No new runtime dependencies added
• Docs updated for behavior changes (README.md)

[dezgit2025]

Addressed via #11 — took the multi-storage concept but reimplemented with security hardening and provider isolation. Thank you for the contribution and the idea, @jshessen! The provider architecture, trust fencing, bounded JSONL reads, and token budget enforcement in v0.2.0 were all inspired by your work here.

[dezgit2025]

Hey @jshessen — wanted to give you proper credit here. Your PR laid the groundwork for the multi-storage provider architecture we shipped in v0.2.0. The provider-based fallback design, the VS Code/JetBrains/Neovim backend concepts, and the JSONL event parsing all trace back to your work on this PR.

We ended up refactoring significantly before merging (different module structure, tighter schema validation, some scope changes), but the core idea of pluggable storage providers originated here. You've been added to the Contributors section in the README. Thanks for the solid foundation. 🙏

[jshessen]

Hey @jshessen — wanted to give you proper credit here. Your PR laid the groundwork for the multi-storage provider architecture we shipped in v0.2.0. The provider-based fallback design, the VS Code/JetBrains/Neovim backend concepts, and the JSONL event parsing all trace back to your work on this PR.

We ended up refactoring significantly before merging (different module structure, tighter schema validation, some scope changes), but the core idea of pluggable storage providers originated here. You've been added to the Contributors section in the README. Thanks for the solid foundation. 🙏

Perfect -- I am just glad it can be extended to cover the other use cases