delimit-cli 4.6.0 → 4.6.2

package/gateway/ai/ledger_manager.py

@@ -124,7 +124,16 @@ def _detect_venture(project_path: str = ".") -> Dict[str, str]:
 
 
 def _register_venture(info: Dict[str, str]):
-    """Silently register a venture in the global registry."""
+    """Silently register a venture in the global registry.
+
+    Phase C follow-up (2026-05-18): reject paths under /tmp/* or the
+    bare "/tmp" itself. Pytest tmp_path values leaked into the registry
+    as ventures (`tmp: /tmp`, `test_project: /tmp/pytest-of-root/...`),
+    causing every fresh tmp_path to match via path-prefix in
+    resolve_venture and breaking test_resolve_venture_unregistered_path.
+    The guard fails-silently — tests that pass tmp_path to functions
+    which auto-register simply don't pollute the registry going forward.
+    """
     GLOBAL_DIR.mkdir(parents=True, exist_ok=True)
     ventures = {}
     if VENTURES_FILE.exists():
@@ -134,9 +143,19 @@ def _register_venture(info: Dict[str, str]):
             pass
 
     name = info["name"]
+    path = info.get("path", "")
+    # Guard against the specific test-state pollution that broke
+    # test_resolve_venture_unregistered_path: a `tmp: /tmp` venture
+    # caught EVERY pytest tmp_path via path-prefix in resolve_venture.
+    # Reject bare "/tmp" only. Deeper /tmp/<X> paths are fine — they
+    # only path-prefix-match their own subtree, not every tmp_path,
+    # AND legitimate test fixtures (e.g. test_ledger_proof) register
+    # subpaths during a single test run and need that to work.
+    if path == "/tmp" or path.rstrip("/") == "/tmp":
+        return
     if name not in ventures:
         ventures[name] = {
-            "path": info.get("path", ""),
+            "path": path,
             "repo": info.get("repo", ""),
             "type": info.get("type", ""),
             "registered_at": time.strftime("%Y-%m-%dT%H:%M:%SZ"),
@@ -565,8 +584,18 @@ def update_item(
     blocks: Optional[str] = None,
     project_path: str = ".",
     worked_by: str = "",
+    commit_sha: Optional[str] = None,
+    pr_url: Optional[str] = None,
 ) -> Dict[str, Any]:
-    """Update an existing ledger item's fields."""
+    """Update an existing ledger item's fields.
+
+    LED-1408 Phase 1: when `status="done"` is requested, callers MAY provide
+    `commit_sha` and/or `pr_url` as proof that the work shipped to main.
+    The proof is recorded on the update event under `ship_proof` with a
+    `verified: bool` flag. Phase 1 does NOT enforce — items still
+    transition to `done` even without proof — but the flag lets future
+    audits and the Phase 2 reconciler find unverified-done items.
+    """
     _ensure(project_path)
     ledger_dir = _project_ledger_dir(project_path)
 
@@ -633,6 +662,49 @@ def update_item(
             update["blocked_by"] = blocked_by
         if blocks:
             update["blocks"] = blocks
+
+        # LED-1408 Phase 1: attach ship_proof block when status transitions to
+        # `done` or `shipped_pending`. Verified=True iff commit_sha or pr_url
+        # was supplied (directly or scraped from the note). Phase 2's
+        # reconciler will use this to distinguish "trustworthy done" from
+        # "marked done but never verified on main."
+        if status in ("done", "shipped_pending"):
+            try:
+                from ai.ledger_proof import build_ship_proof
+                update["ship_proof"] = build_ship_proof(
+                    commit_sha=commit_sha,
+                    pr_url=pr_url,
+                    note=note,
+                )
+                # LED-1420 Phase 2 strict-mode flip: when DELIMIT_LEDGER_STRICT_DONE=1,
+                # an unverified `done` transition is downgraded to `shipped_pending`
+                # so the nightly reconciler (scripts/delimit_ledger_reconciler.py)
+                # can promote it to `done` once a commit-trailer match shows up on
+                # origin/main. Off by default so existing workflows keep closing
+                # items without hitting an unexpected gate; flip when the
+                # reconciler has been observed running for ~1 week without
+                # surprises.
+                strict = os.environ.get("DELIMIT_LEDGER_STRICT_DONE") == "1"
+                if (
+                    strict
+                    and status == "done"
+                    and not update["ship_proof"].get("verified")
+                ):
+                    update["status"] = "shipped_pending"
+                    existing_note = update.get("note") or ""
+                    suffix = (
+                        "[LED-1420 strict-mode: downgraded done → shipped_pending — "
+                        "no commit_sha/pr_url proof; reconciler will upgrade to "
+                        "done when it finds a Ledger-Item: " + item_id + " trailer "
+                        "on origin/main]"
+                    )
+                    update["note"] = (existing_note + " " + suffix).strip() if existing_note else suffix
+            except Exception:
+                # Soft-fail: a ship_proof bug must not break ledger close.
+                # The unverified state will be re-detectable from the missing
+                # key on the next audit pass.
+                pass
+
         _append(path, update)
 
         # Sync to Supabase for dashboard visibility
@@ -1306,7 +1378,12 @@ def session_history(limit: int = 5) -> Dict[str, Any]:
 # `archive` is a soft transition (status="archived", appended to JSONL); items
 # stay in replay forever. NO hard delete. Per-item failures don't block others.
 BULK_ACTIONS = ("archive", "set_status", "set_priority", "add_tag", "mark_done", "cancel")
-_VALID_BULK_STATUSES = ("open", "in_progress", "blocked", "done", "cancelled", "archived", "completed")
+# LED-1408: `shipped_pending` is the intermediate state between "committed" and
+# "verified on main." Items transition to shipped_pending when a worker reports
+# completion (commit exists somewhere) but the orchestrator hasn't yet verified
+# the commit is reachable from origin/main. The reconciler (Phase 2) promotes
+# shipped_pending → done once reachability is confirmed.
+_VALID_BULK_STATUSES = ("open", "in_progress", "blocked", "shipped_pending", "done", "cancelled", "archived", "completed")
 _VALID_BULK_PRIORITIES = ("P0", "P1", "P2", "P3")
 
 

package/gateway/ai/ledger_proof.py

@@ -0,0 +1,127 @@
+"""Ledger ship-state proof helpers (LED-1408 Phase 1).
+
+When a ledger item transitions to `done`, we want auditable evidence that the
+fix actually shipped. Two forms of proof:
+
+1. **Commit-trailer binding** — the merge commit carries a `Ledger-Item:
+   LED-NNNN` trailer. Lets a webhook or reconciler walk `git log
+   origin/main` and find every commit-to-ledger link without naming
+   conventions or fuzzy matching.
+
+2. **PR-URL linkage** — `https://github.com/<org>/<repo>/pull/<N>` plus a
+   verified `merged_at` timestamp. Fallback for items closed without a
+   trailer.
+
+Phase 1 (this module): parse + record. Items closed with proof get
+`verified: true` on the event; items closed without proof get
+`verified: false` so a future audit can find them.
+
+Phase 2 (separate LED): the reconciler enforces stricter semantics —
+items without proof default to `shipped_pending`, not `done`.
+
+Memory anchor: feedback_agent_dashboard_done_means_committed_not_merged.md
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Dict, Optional
+
+# Match `Ledger-Item: LED-1234` (case-insensitive, leading whitespace allowed).
+# Pattern intentionally tolerant of trailing whitespace + multiple LED IDs:
+# we extract the FIRST LED-N on the line.
+_LEDGER_TRAILER_RE = re.compile(
+    r"(?im)^\s*Ledger-Item\s*:\s*(LED-\d+)",
+)
+
+# Match a GitHub PR URL: https://github.com/<owner>/<repo>/pull/<N>
+_PR_URL_RE = re.compile(
+    r"https://github\.com/([\w.-]+)/([\w.-]+)/pull/(\d+)",
+)
+
+
+def parse_ledger_trailer(commit_message: str) -> Optional[str]:
+    """Extract the `Ledger-Item: LED-NNNN` trailer value from a commit message.
+
+    Returns the LED id (e.g. `LED-1408`) or None if no trailer is present.
+    The trailer must be on its own line; mentions inside prose (e.g.
+    `mentions LED-1408 in passing`) do NOT match.
+    """
+    if not commit_message:
+        return None
+    match = _LEDGER_TRAILER_RE.search(commit_message)
+    if match:
+        return match.group(1)
+    return None
+
+
+def parse_pr_url(text: str) -> Optional[Dict[str, str]]:
+    """Extract the first GitHub PR URL from any string.
+
+    Returns {owner, repo, number} or None.
+    """
+    if not text:
+        return None
+    match = _PR_URL_RE.search(text)
+    if match:
+        return {
+            "owner": match.group(1),
+            "repo": match.group(2),
+            "number": match.group(3),
+        }
+    return None
+
+
+def build_ship_proof(
+    commit_sha: Optional[str] = None,
+    pr_url: Optional[str] = None,
+    note: Optional[str] = None,
+) -> Dict[str, object]:
+    """Build a ship-proof block to attach to a ledger `done` event.
+
+    Inputs may come from explicit MCP-tool args OR from inline mentions
+    in the note (the caller might paste a PR URL into the note field
+    without realizing it's also queryable).
+
+    Returns a dict with keys:
+      - verified: bool — True iff commit_sha OR pr_url was provided
+      - commit_sha: str or None
+      - pr_url: str or None
+      - pr_owner / pr_repo / pr_number: str or None (parsed from pr_url)
+      - ledger_trailer: str or None (parsed from note, if present)
+
+    The `verified` flag is the primary downstream signal. A future
+    reconciler will refuse to transition `done` without verified=True;
+    Phase 1 only records the flag without enforcing.
+    """
+    proof: Dict[str, object] = {
+        "verified": bool(commit_sha or pr_url),
+        "commit_sha": commit_sha or None,
+        "pr_url": pr_url or None,
+    }
+
+    if pr_url:
+        parsed = parse_pr_url(pr_url)
+        if parsed:
+            proof["pr_owner"] = parsed["owner"]
+            proof["pr_repo"] = parsed["repo"]
+            proof["pr_number"] = parsed["number"]
+
+    # If pr_url not explicitly passed but appears in the note, capture it
+    if not pr_url and note:
+        parsed = parse_pr_url(note)
+        if parsed:
+            proof["pr_url"] = f"https://github.com/{parsed['owner']}/{parsed['repo']}/pull/{parsed['number']}"
+            proof["pr_owner"] = parsed["owner"]
+            proof["pr_repo"] = parsed["repo"]
+            proof["pr_number"] = parsed["number"]
+            proof["verified"] = True
+
+    # Capture any Ledger-Item trailer from the note (rare but possible —
+    # a worker might paste the commit message into the close note).
+    if note:
+        trailer = parse_ledger_trailer(note)
+        if trailer:
+            proof["ledger_trailer"] = trailer
+
+    return proof

package/gateway/ai/license.py

@@ -20,12 +20,77 @@ try:
         PRO_TOOLS as _CORE_PRO_TOOLS,
         FREE_TRIAL_LIMITS,
     )
-    # Extend compiled PRO_TOOLS with tools added after last binary build
+    # Extend compiled PRO_TOOLS with tools added after last binary build.
+    # LED-1260: keep this in lockstep with the fallback set below — any tool
+    # in the fallback PRO_TOOLS that's NOT in the compiled set must be added
+    # here, otherwise customers with the binary get those tools FREE while
+    # customers without the binary pay for them (regression-on-success).
+    # The runtime test in tests/test_license.py asserts both sets are
+    # equal. LED-1410 makes this stronger: the extension set below is
+    # CODEGEN from ai/pro_tools.yaml (same SSoT as the compiled
+    # set), so the two are equal by construction. The | union with
+    # _CORE_PRO_TOOLS is preserved so OLDER compiled .so files that
+    # were built before a YAML addition still pick up the new tool
+    # at runtime.
     PRO_TOOLS = _CORE_PRO_TOOLS | frozenset({
-        "delimit_social_approve",
-        # Autonomous build loop
-        "delimit_next_task", "delimit_task_complete",
-        "delimit_loop_status", "delimit_loop_config",
+        # CODEGEN-START: EXTENSION_PRO_TOOLS
+    "delimit_agent_complete",
+    "delimit_agent_dispatch",
+    "delimit_agent_handoff",
+    "delimit_agent_status",
+    "delimit_cost_alert",
+    "delimit_cost_analyze",
+    "delimit_cost_optimize",
+    "delimit_deliberate",
+    "delimit_deploy_build",
+    "delimit_deploy_npm",
+    "delimit_deploy_plan",
+    "delimit_deploy_publish",
+    "delimit_deploy_rollback",
+    "delimit_deploy_site",
+    "delimit_deploy_status",
+    "delimit_deploy_verify",
+    "delimit_evidence_collect",
+    "delimit_evidence_verify",
+    "delimit_executor",
+    "delimit_gov_evaluate",
+    "delimit_gov_new_task",
+    "delimit_gov_policy",
+    "delimit_gov_run",
+    "delimit_gov_verify",
+    "delimit_loop_config",
+    "delimit_loop_status",
+    "delimit_memory_search",
+    "delimit_models",
+    "delimit_next_task",
+    "delimit_notify",
+    "delimit_obs_logs",
+    "delimit_obs_metrics",
+    "delimit_obs_status",
+    "delimit_os_gates",
+    "delimit_os_plan",
+    "delimit_os_status",
+    "delimit_release_plan",
+    "delimit_release_status",
+    "delimit_release_sync",
+    "delimit_repo_analyze",
+    "delimit_repo_config_audit",
+    "delimit_repo_config_validate",
+    "delimit_repo_diagnose",
+    "delimit_screen_record",
+    "delimit_screenshot",
+    "delimit_security_deliberate",
+    "delimit_security_ingest",
+    "delimit_social_approve",
+    "delimit_social_generate",
+    "delimit_social_history",
+    "delimit_social_post",
+    "delimit_task_complete",
+    "delimit_test_coverage",
+    "delimit_vault_health",
+    "delimit_vault_search",
+    "delimit_vault_snapshot",
+        # CODEGEN-END: EXTENSION_PRO_TOOLS
     })
 except ImportError:
     # license_core not available — three known cases:
@@ -50,49 +115,69 @@ except ImportError:
 
     LICENSE_FILE = Path.home() / ".delimit" / "license.json"
 
+    # LED-1410: CODEGEN from ai/pro_tools.yaml — same SSoT as the
+    # compiled set above. Memory note preserved here for source readers:
+    # delimit_memory_store + delimit_memory_recent are FREE (LED-193).
+    # Only delimit_memory_search is Pro.
     PRO_TOOLS = frozenset({
-        # Governance deep
-        "delimit_gov_evaluate", "delimit_gov_policy", "delimit_gov_run", "delimit_gov_verify",
-        "delimit_gov_new_task",
-        # OS layer
-        "delimit_os_plan", "delimit_os_status", "delimit_os_gates",
-        # Deploy pipeline
-        "delimit_deploy_plan", "delimit_deploy_build", "delimit_deploy_publish",
-        "delimit_deploy_verify", "delimit_deploy_rollback", "delimit_deploy_status",
-        "delimit_deploy_site", "delimit_deploy_npm",
-        # Memory (search is Pro; store + recent are free)
-        "delimit_memory_search",
-        "delimit_vault_search", "delimit_vault_snapshot", "delimit_vault_health",
-        # Evidence
-        "delimit_evidence_collect", "delimit_evidence_verify",
-        # Deliberation + Models
-        "delimit_deliberate", "delimit_models",
-        # Security orchestrator
-        "delimit_security_ingest", "delimit_security_deliberate",
-        # Observability
-        "delimit_obs_metrics", "delimit_obs_logs", "delimit_obs_status",
-        # Release
-        "delimit_release_plan", "delimit_release_status", "delimit_release_sync",
-        # Cost
-        "delimit_cost_analyze", "delimit_cost_optimize", "delimit_cost_alert",
-        # Social
-        "delimit_social_post", "delimit_social_generate", "delimit_social_history",
-        "delimit_social_approve",
-        # Repo deep
-        "delimit_repo_analyze", "delimit_repo_config_audit", "delimit_repo_config_validate",
-        "delimit_repo_diagnose",
-        # Test
-        "delimit_test_coverage",
-        # Screen recording
-        "delimit_screen_record", "delimit_screenshot",
-        # Notifications
-        "delimit_notify",
-        # Agent orchestration
-        "delimit_agent_dispatch", "delimit_agent_status",
-        "delimit_agent_complete", "delimit_agent_handoff",
-        # Autonomous build loop
-        "delimit_next_task", "delimit_task_complete",
-        "delimit_loop_status", "delimit_loop_config",
+        # CODEGEN-START: FALLBACK_PRO_TOOLS
+    "delimit_agent_complete",
+    "delimit_agent_dispatch",
+    "delimit_agent_handoff",
+    "delimit_agent_status",
+    "delimit_cost_alert",
+    "delimit_cost_analyze",
+    "delimit_cost_optimize",
+    "delimit_deliberate",
+    "delimit_deploy_build",
+    "delimit_deploy_npm",
+    "delimit_deploy_plan",
+    "delimit_deploy_publish",
+    "delimit_deploy_rollback",
+    "delimit_deploy_site",
+    "delimit_deploy_status",
+    "delimit_deploy_verify",
+    "delimit_evidence_collect",
+    "delimit_evidence_verify",
+    "delimit_executor",
+    "delimit_gov_evaluate",
+    "delimit_gov_new_task",
+    "delimit_gov_policy",
+    "delimit_gov_run",
+    "delimit_gov_verify",
+    "delimit_loop_config",
+    "delimit_loop_status",
+    "delimit_memory_search",
+    "delimit_models",
+    "delimit_next_task",
+    "delimit_notify",
+    "delimit_obs_logs",
+    "delimit_obs_metrics",
+    "delimit_obs_status",
+    "delimit_os_gates",
+    "delimit_os_plan",
+    "delimit_os_status",
+    "delimit_release_plan",
+    "delimit_release_status",
+    "delimit_release_sync",
+    "delimit_repo_analyze",
+    "delimit_repo_config_audit",
+    "delimit_repo_config_validate",
+    "delimit_repo_diagnose",
+    "delimit_screen_record",
+    "delimit_screenshot",
+    "delimit_security_deliberate",
+    "delimit_security_ingest",
+    "delimit_social_approve",
+    "delimit_social_generate",
+    "delimit_social_history",
+    "delimit_social_post",
+    "delimit_task_complete",
+    "delimit_test_coverage",
+    "delimit_vault_health",
+    "delimit_vault_search",
+    "delimit_vault_snapshot",
+        # CODEGEN-END: FALLBACK_PRO_TOOLS
     })
     FREE_TRIAL_LIMITS = {"delimit_deliberate": 3}
 

package/gateway/ai/license_core.pyi

@@ -7,7 +7,7 @@ import os
 from pathlib import Path
 import hashlib
 
-PRO_TOOLS = frozenset({'delimit_gov_evaluate', 'delimit_gov_policy', 'delimit_gov_run', 'delimit_gov_verify', 'delimit_os_plan', 'delimit_os_status', 'delimit_os_gates', 'delimit_deploy_plan', 'delimit_deploy_build', 'delimit_deploy_publish', 'delimit_deploy_verify', 'delimit_deploy_rollback', 'delimit_deploy_status', 'delimit_deploy_site', 'delimit_deploy_npm', 'delimit_memory_search', 'delimit_vault_search', 'delimit_vault_snapshot', 'delimit_vault_health', 'delimit_evidence_collect', 'delimit_evidence_verify', 'delimit_deliberate', 'delimit_models', 'delimit_obs_metrics', 'delimit_obs_logs', 'delimit_obs_status', 'delimit_release_plan', 'delimit_release_status', 'delimit_release_sync', 'delimit_cost_analyze', 'delimit_cost_optimize', 'delimit_cost_alert', 'delimit_social_post', 'delimit_social_generate', 'delimit_social_history', 'delimit_screen_record', 'delimit_screenshot', 'delimit_notify', 'delimit_agent_dispatch', 'delimit_agent_status', 'delimit_agent_complete', 'delimit_agent_handoff', 'delimit_executor'})
+PRO_TOOLS = frozenset({'delimit_agent_complete', 'delimit_agent_dispatch', 'delimit_agent_handoff', 'delimit_agent_status', 'delimit_cost_alert', 'delimit_cost_analyze', 'delimit_cost_optimize', 'delimit_deliberate', 'delimit_deploy_build', 'delimit_deploy_npm', 'delimit_deploy_plan', 'delimit_deploy_publish', 'delimit_deploy_rollback', 'delimit_deploy_site', 'delimit_deploy_status', 'delimit_deploy_verify', 'delimit_evidence_collect', 'delimit_evidence_verify', 'delimit_executor', 'delimit_gov_evaluate', 'delimit_gov_policy', 'delimit_gov_run', 'delimit_gov_verify', 'delimit_memory_search', 'delimit_models', 'delimit_notify', 'delimit_obs_logs', 'delimit_obs_metrics', 'delimit_obs_status', 'delimit_os_gates', 'delimit_os_plan', 'delimit_os_status', 'delimit_release_plan', 'delimit_release_status', 'delimit_release_sync', 'delimit_screen_record', 'delimit_screenshot', 'delimit_social_generate', 'delimit_social_history', 'delimit_social_post', 'delimit_vault_health', 'delimit_vault_search', 'delimit_vault_snapshot'})
 def needs_revalidation(data: dict) -> bool:
     ...
 def revalidate_license(data: dict) -> dict: