Commit Graph

7 Commits

Author SHA1 Message Date
Dai Ha f9073e2320 docs: CB-301-ext spec — worktree provisioning + config-parity overlay
Opt-in per-acquire worktree (shared-tree default preserved). git behind a
Worktrees seam (ProcessBuilder impl, fake in tests). acquire provisions
worktree+branch, overlays local config (copy + --skip-worktree on tracked
files so the worker can't commit .mcp.json), spawns with cwd=worktree.
release removes the worktree but keeps the branch (holds commits/PR).
WorkerSession gains nullable worktree/branch; BridgedConfig.Worker gains
parityOverlay. 6 fake-based acceptance tests incl. backward-compat + unwind.
2026-07-17 06:32:09 +02:00
Dai Ha 82c7d6553a docs: worker git workflow — daemon worktree + config parity + worker-opened PR
Worktree is code-only isolation; SessionManager hydrates it to full config
parity (overlay untracked local settings/.mcp.json/.env) so a worker is a
full peer of the primary, differing only in the LLM provider. Worker commits,
pushes over SSH, and opens its own PR (gitea REST + repo-scoped token).
2026-07-16 19:25:41 +02:00
Dai Ha 19b10e3216 docs: CB-301 session-manager design spec (one-shot, no reuse; recycle in scope)
The as-built audit surfaced that WorkerService keeps no registry of what it
spawned (its own Javadoc: 'there is no registry; list() only asks herdr').
CB-301 adds a SessionManager wrapping WorkerService: an authoritative in-daemon
roster with a per-session lifecycle FSM (SPAWNING/READY/BUSY/DONE/RELEASED/
FAILED), deterministic release, and recycle (= release + fresh acquire, no
reuse). Leaves clean seams for CB-302 (checkpoint on release), CB-303 (idle_ttl/
context_cap/drain policy over roster), CB-304 (bridge_list reads roster).
2026-07-16 19:10:13 +02:00
Dai Ha 926724a279 CB-112: workers inherit the primary's working directory (not $HOME)
A worker now opens the same directory the primary is in, unless told otherwise.
Resolution: explicit spawn cwd → per-profile config cwd → the primary's cwd
(auto-detected from the bridge_spawn caller's PID via lsof -d cwd) → the daemon's
cwd. Never $HOME.

Mechanism (found by live probe, corrects the earlier assumption): an agent.start
pane does NOT inherit its tab's or workspace's cwd — it starts in $HOME. herdr's
agent.start honours an (undocumented) cwd param, so the resolved cwd is threaded
onto agent.start {cwd} (both tab and pane placement), not tab.create.

Surfaces: bridge_spawn {cwd?} + auto-detect via ConnectionIdentity.resolve (peer
PID) + ProcessCwdLookup (lsof); REST POST /workers ?cwd= / body cwd; per-profile
'cwd:' config. Validated live: explicit cwd → worker rooted there; no cwd over
REST → daemon cwd, not $HOME. Also clears the ccs folder-trust prompt when the
project dir is already trusted (see docs/Worker-Startup-and-Trust.md).
2026-07-15 16:33:46 +02:00
Dai Ha 959f04bc96 docs: worker startup — working directory & the folder-trust prompt
Documents (a) the rule that a worker inherits the PRIMARY's working directory,
never $HOME — resolved from an explicit cwd, else the bridge_spawn caller's PID
(lsof -d cwd), else the daemon cwd; herdr's seam is workspace.create {cwd}, since
agent.start has no cwd; (b) ccs (Claude Code) as the assumed launcher and its
folder-trust model (hasTrustDialogAccepted per project in each instance's
.claude.json), so trusting the project dir once per profile clears the prompt;
(c) that other CLIs have their own startup gates, documented per launcher.
Marks the cwd-inheritance as target design (follow-up), not yet wired.
2026-07-15 16:11:18 +02:00
kevin c7b58f2195 docs: add Team.md under docs/
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-14 19:14:25 +07:00
kevin 01cfda7965 Add MCP design doc 2026-07-14 12:30:20 +07:00