The data-show-github command can fail in CI when no seeded data
exists. Mark it allowFailure: true. Also update the log path
normalizer regex for ~/.vana (was ~/.dataconnect).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

These commands fail in CI where no collected data or browser exists.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…cripts

Update VHS fixture directory, transcript paths, all documentation
references, and utility scripts to use ~/.vana consistently.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Skill list: one line per skill, "installed" badge in accent blue.
No descriptions (use vana skill show for detail). Matches the
compact list pattern.

Skill install: success line + next step only. No file paths.
Suggests installing the next uninstalled skill instead of
circular vana skill list.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Async/background connect, scheduled collection, SEA stack trace,
stale transcripts, date filtering, data versioning.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Stale SingletonLock after interrupted connect, agent-friendly
credential passing, MCP auth mode check, skill composition
boundaries.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add IPC mode issue: expose existing file-based requestInput IPC
(pending-input/input-response JSON files) as a CLI mode so agents
can handle credential prompts without interactive stdin.

connect-data skill: check auth mode first, run interactive sources
in foreground (not background), tell user to run legacy sources
themselves.

next-prompt skill: never attempt connections, work with available
data only.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

IPC mode (--ipc): connectors pause and write credential questions
to ~/.vana/pending-input-{runId}.json, then poll for the agent's
response at ~/.vana/input-response-{runId}.json. Enables agents
to handle auth without interactive stdin. 5-minute timeout.

Stale lock cleanup: remove SingletonLock before every browser
launch, preventing "Failed to create ProcessSingleton" errors
after interrupted connects.

MCP auth check: connect_source tool checks authMode before
spawning. Legacy (browser) sources return immediately with
instructions for the user instead of hanging.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The skill now tells agents to use `vana connect <source> --json --ipc`
which writes credential prompts to files instead of stdin. Agents
poll for pending-input files, ask the user, and write response files.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The agent MUST use run_in_background for --ipc connects, then
immediately read the task output for needs-input events. Explicit
step-by-step flow covering background start, output reading,
credential collection, response writing, and 2FA re-prompts.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

30 minutes gives the agent enough time to ask the user for
credentials and handle multi-step auth (2FA). Previously 5
minutes caused timeouts during normal agent-user interaction.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Three-round design analysis (brainstorm > design > plan > assess)
informed by 68 findings across 11 reference CLIs. Conclusion:
blocking is correct for <5 minute operations. No --detach for v1.

next-prompt skill: added freshness check — suggests recollection
when data is >24h old. collect --all already exists.

Research artifacts in research/async-cli/.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

--detach: spawns a background process for long-running connects
(ChatGPT takes 30 minutes). Parent exits immediately, child
writes progress to ~/.vana/sessions/. vana status shows active
background connections.

Connection health: each source tracks healthy/needs_reauth/error/
stale based on actual collection results. vana status shows health
per source with colored indicators. When a source needs re-auth,
status prominently surfaces it.

vana schedule: OS-native scheduling via launchd (macOS) or crontab
(Linux). schedule add --every 24h installs a recurring collection.
schedule list shows active schedules. schedule remove uninstalls.
No custom daemon — the OS is the scheduler.

Design informed by 105 research findings across Plaid, Stripe, MX,
PM2, Turborepo, Docker, Vercel, Railway, and 7 other reference
CLIs/services. Three-round design analysis in
research/async-cli/daemon-design-rounds.md.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Auth expiry: Plaid three-tier health model, advance warning webhooks,
MX 13+ connection statuses, Google silent revocation triggers,
Strava rotating refresh tokens, cross-platform re-auth UX patterns.

Daemon patterns: PM2 dual-socket architecture, Turborepo on-demand
daemon with graceful degradation, Docker socket activation, Homebrew
services OS wrapper, Bree/Croner job schedulers.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Detached processes have no stdin and nobody watching. If auth is
needed, fail fast and record needs_reauth (Plaid's pattern: fail
the fetch, record state, let user re-auth at convenience).

Pre-check: refuse to detach if source has never been connected.
"Run vana connect <source> first to authenticate."

--ipc remains separate for agents actively watching (Claude Code
background task flow). --detach is for fire-and-forget.

Design analysis in research/async-cli/detach-ipc-design.md.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Homebrew formula, hosted installer, and demo assets can go stale
when canary releases don't trigger downstream sync. Document the
full distribution chain that needs automation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Homebrew: trigger formula sync from CI after verify-prerelease-install.
Eliminates the race window where binaries update before the formula,
causing checksum mismatch on brew install.

Error results: detect connector results that are error objects
(e.g. {"error":"Could not determine username"}) and report failure
instead of declaring success.

Sources list: show all sources with descriptions instead of
collapsing after 3. Users were confused by the collapsed line.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add checkLegacyDataHome() that detects ~/.dataconnect without
auto-migrating. The CLI informs the user and tells them how to
move their data. No silent renames, no side effects in path
getters.

Remove shipped items from open issues: IPC mode, stale lock
cleanup, MCP auth check, skill composition, agent credential
passing, homebrew auto-sync.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

On first run, if ~/.dataconnect exists and ~/.vana doesn't:
rename to ~/.vana and create a symlink at ~/.dataconnect pointing
to ~/.vana. DataConnect (which references ~/.dataconnect) follows
the symlink. Both apps share the same data.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Stripe/Vercel convention: plural nouns for resource commands.
sources, skills, not source, skill.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The list showed display names ("Next Prompt") but install needs
IDs ("next-prompt"). Now shows IDs directly. Guessable commands.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Auth mode detection checked the local data-connectors checkout
which could be stale (showing legacy when connectors were already
modernized to requestInput). Now checks the connector cache first
(populated by vana connect's auto-fetch), falls back to local
checkout only if not cached.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CLI README at cli/README.md with current connect output (heartbeat
spinner, scope manifest), full command table including new commands
(--detach, schedule, skills, mcp, collect), updated source list.

Root README points to cli/README.md. Clean separation for future
repo split.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

When ~/.vana exists but ~/.dataconnect doesn't (new CLI user),
create a symlink so DataConnect desktop app finds the data.
DataConnect hardcodes ~/.dataconnect in connector.rs:66 and
should adopt ~/.vana upstream.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

158 deduplicated findings across auth expiry, agent autonomy,
memory infrastructure, personal AI products, and competitive
landscape. Three artifacts: landscape CSV (40 rows), landscape
report (505 lines), executive memo (72 lines).

Research informs the next-prompt skill and vana next feature
design. Key finding: no product combines cross-platform personal
data with coding agent context.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Remove superseded docs (CLI-README.md, CLI-README-launch.md,
EXECUTION-CONTEXT-MEMO.md). Move docs/research/ to
research/cli-design/ alongside async-cli and personal-data-agents.
Add creation datestamps to 10 docs that were missing them.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

GITHUB_TOKEN lacks permission to dispatch workflows in
vana-com/homebrew-vana. Use a fine-grained PAT with Actions
write scope on that repo.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>