@deftai/directive-content 0.98.1 → 0.100.0

package/templates/agents-entry.md

@@ -83,6 +83,17 @@ Legacy `vbrief/` read-accepted; `deft migrate:xbrief` for `xbrief/` (v0.6→v0.8
 ⊗ `stop-at: pr-open` without a named babysit / merge-path owner, or dual review-monitor leases on recovery (#3044 / #2261).
 ⊗ Stand down at CLEAN under human-merge with no reachable post-merge `scope:complete` owner.
 
+## Nuclear-family A2A topology (#3155)
+
+! Agent-to-agent messaging is **nuclear-family** only: parent / sibling (same cohort) / child. Cross-cohort or cross-session coordination goes through a shared parent or durable parent-owned artifacts — not peer mesh. Depth: `.deft/core/swarm/swarm.md` `## Communication Topology (#3155)`; security: `.deft/core/meta/security.md` `## Unbounded A2A graphs (#3155)`; ADR: `docs/decisions/ADR-003-a2a-nuclear-family-topology.md` (decision input to #2705; client-posture ADR remainder stays on #2705). Pair: retained children #3158; parent epic #3179.
+⊗ Open-mesh agent-to-agent messaging across cohorts or sessions ("agents everywhere").
+⊗ Treat retained / re-addressable children as license to mesh outside the nuclear family.
+
+## Mid-scope gate capability tier (#3158 / #954)
+
+! Mid-scope gates: **split-dispatch** when `agent_id` is terminal; retain-capable hosts (continue-by-agent-id / message-later / steer-mid-flight) MAY re-message the live child. Retention = orchestration only — not constitution self-edit (#3164). Depth: preamble §10; `deft-directive-swarm`. Topology: #3155 nuclear-family. ⊗ Invent retain on one-shot hosts.
+
+
 ## Review-surface precedence (#2308)
 
 ! Route PR shepherding / review work through `deft-directive-review-cycle` — `.deft/core/.agents/skills/deft-directive-review-cycle/SKILL.md`; host `babysit` / `bugbot` / `security-review` advisory-only (#2308 / #2261).
@@ -91,6 +102,9 @@ Legacy `vbrief/` read-accepted; `deft migrate:xbrief` for `xbrief/` (v0.6→v0.8
 
 ! `plan.policy.valueFeedback.enabled` defaults OFF — `deft policy:show --field=valueFeedback` / `deft policy:enable-value-feedback -- --confirm`; `deft value:show`; `deft feedback:file`; `.deft/core/.agents/skills/deft-directive-feedback/SKILL.md` (#1709).
 
+## Structured decision log (#1396 / #3211)
+! Significant choices → `deft decision:write`; re-load → `deft decision:list` / `xbrief/decisions/`; depth `.deft/core/docs/decision-log.md` (not triage/ADRs/lessons).
+
 ## Eval and framework health (#1703)
 
 ! `deft eval:health` when orienting or after gate/policy changes (Tier 0; 4-hour debounce). Release: `deft eval:run` / `deft eval:report`; skill routing: `deft eval:triggers` (#1586 / #1703).
@@ -114,6 +128,10 @@ Legacy `vbrief/` read-accepted; `deft migrate:xbrief` for `xbrief/` (v0.6→v0.8
 
 ## Development Process
 
+### Gate integrity (#3156)
+
+! When a quality gate fails, fix the product/process/test under test — ⊗ clear red by editing the gate definition, verifier, reward, required check, coverage floor, or policy flag solely to go green. Deliberate gate changes go through issue/PR + review. Depth: `.deft/core/docs/gate-integrity.md` (refine-internal SkillOpt stays on #2436).
+
 ### Implementation Intent Gate (#810 / #1193)
 
 ! `deft xbrief:preflight -- <path>` on `xbrief/active/` before code-writing; action-verb (`build`, `implement`, `ship`, `swarm`, `run agents`, `start agent`) (#810). Slash-command sessions inherit only that verb (`DEFT_SESSION_SLASH_VERB`); non-implement verbs (`/github-issue`, `/triage`, …) MUST NOT authorize implement/push/PR/merge/deploy (#1193) — `commands.md` / `contracts/intent-ceiling.md`.

package/templates/swarm-greptile-poller-prompt.md

@@ -196,20 +196,64 @@ for m in _TIER25_RE.finditer(body):
 # Greptile sometimes inlines the verdict as plain prose, e.g.
 #     Three P1 findings (two from prior review, one new): wrong exception ...
 #     Not safe to merge until the mocked-import test defect is resolved.
+#     should-not-merge / Do not merge until residual risk is documented (#3225)
+#     Safe to merge once corrected
 #     P1 -- wrong exception type for state validation in populate()
 # Negation-context guard applies to the count-prose sentinel (`No P0 findings`,
-# `Zero P1 findings` MUST NOT trigger). The `Not safe to merge` substring is
-# Greptile's explicit human-readable verdict and is treated as a hard block.
+# `Zero P1 findings` MUST NOT trigger). Advisory should-not-merge prose is a
+# hard block even when formal review is still Comment (#3225); composes with
+# minGreptileConfidence (#3095). Canonical TS detector:
+# packages/core/src/content-contracts/skills/greptile-detector.ts
 _TIER3_COUNT_RE = re.compile(
     r"\b(?:One|Two|Three|Four|Five|Six|Seven|Eight|Nine|Ten|\d+)\s+P[01]\s+findings?\b",
     re.IGNORECASE,
 )
 _TIER3_LINE_RE = re.compile(r"^\s*P[01]\s+--\s", re.MULTILINE)
 _TIER3_NEGATIONS = ("No ", "Zero ", "no ", "NO ")
+# #3225 advisory should-not-merge family — LINE-ANCHORED (parity with
+# packages/core/src/content-contracts/skills/greptile-detector.ts).
+# Mid-sentence descriptive mentions ("adds should-not-merge matching") MUST NOT
+# trigger; Summary:/Decision: labels, bullets, and short subject prefixes
+# ("The PR is …") are stripped before the match.
+_ADVISORY_SHOULD_NOT_MERGE_RES = (
+    re.compile(r"\bnot\s+(?:yet\s+)?safe\s+to\s+merge\b", re.IGNORECASE),
+    re.compile(r"\bshould\s*[-–—]?\s*not\s*[-–—]?\s*merge\b", re.IGNORECASE),
+    re.compile(r"\bsafe\s+to\s+merge\s+once\s+corrected\b", re.IGNORECASE),
+    re.compile(r"\bdo\s+not\s+merge\b", re.IGNORECASE),
+    re.compile(r"\bnot\s+ready\s+to\s+merge\b", re.IGNORECASE),
+    re.compile(r"\bnot\s+ready\s+for\s+merge\b", re.IGNORECASE),
+)
+_ADVISORY_LINE_PREFIX_RE = re.compile(
+    r"^(?:Summary|Decision|Verdict)\s*[:\-–—]\s*",
+    re.IGNORECASE,
+)
+_ADVISORY_SUBJECT_RE = re.compile(
+    r"^(?:the\s+pr|this\s+pr|this\s+change|the\s+change|this\s+diff)\s+is\s+",
+    re.IGNORECASE,
+)
+
+def _line_has_anchored_advisory(line: str) -> bool:
+    bare = line.strip()
+    bare = re.sub(r"^>\s*", "", bare)
+    bare = _ADVISORY_LINE_PREFIX_RE.sub("", bare, count=1)
+    bare = re.sub(r"^(?:[-*•]\s+)+", "", bare)
+    bare = re.sub(r"^\*\*", "", bare)
+    bare = re.sub(r"\*\*$", "", bare)
+    bare = re.sub(r"^_", "", bare)
+    bare = re.sub(r"_$", "", bare)
+    bare = _ADVISORY_SUBJECT_RE.sub("", bare, count=1).strip()
+    if not bare:
+        return False
+    for pat in _ADVISORY_SHOULD_NOT_MERGE_RES:
+        m = pat.search(bare)
+        if m is not None and m.start() <= 4:
+            return True
+    return False
 
 def _has_tier3_sentinel(body: str) -> bool:
-    if "Not safe to merge" in body:
-        return True
+    for line in body.splitlines():
+        if _line_has_anchored_advisory(line):
+            return True
     for m in _TIER3_COUNT_RE.finditer(body):
         line = _line_for(body, m.start())
         if any(neg in line for neg in _TIER3_NEGATIONS):

package/vbrief/schemas/vbrief-core.schema.json

@@ -509,6 +509,11 @@
             "critical"
           ]
         },
+        "effort": {
+          "type": "string",
+          "enum": ["S", "M", "L", "XL"],
+          "description": "Optional effort estimate with time anchors: S <2h, M half-day (2-4h), L 1-2 days, XL needs breakdown into S/M/L before activation. Omitted items still validate."
+        },
         "dueDate": {
           "$ref": "#/$defs/dateTime"
         },

package/vbrief/schemas/xbrief-core-0.8.schema.json

@@ -318,6 +318,11 @@
             "critical"
           ]
         },
+        "effort": {
+          "type": "string",
+          "enum": ["S", "M", "L", "XL"],
+          "description": "Optional effort estimate with time anchors: S <2h, M half-day (2-4h), L 1-2 days, XL needs breakdown into S/M/L before activation. Omitted items still validate."
+        },
         "dueDate": {
           "$ref": "#/$defs/dateTime"
         },

package/vbrief/vbrief.md

@@ -419,6 +419,43 @@ Scope vBRIEFs use a small set of **canonical narrative keys** at the `plan.narra
 }
 ```
 
+### Effort estimate (S/M/L/XL) (#1581)
+
+`PlanItem.effort` is an **optional** sizing field on each plan item. Existing items that omit it still validate.
+
+| Value | Time anchor | Meaning |
+|-------|-------------|---------|
+| `S` | less than 2 hours | Safe single-agent slice |
+| `M` | half-day (2–4 hours) | Single-agent; modest scope |
+| `L` | 1–2 days | May warrant a dedicated agent or parallel worktrees |
+| `XL` | needs breakdown | **Must not enter `active/` / `running`** until broken into S/M/L items |
+
+#### Authority and ordering (compose with ceremony dial #3214)
+
+Plan-item `effort` is **post-planning** authority. Estimates live on scope plan items after planning has produced those items. Session-start ceremony (ritual, deposit load, gate setup) runs **before** planning, so plan-item effort is **not** available as session-start ritual input.
+
+- ! Treat `PlanItem.effort` as the planning-time confirm/correct signal for provisional intake size — not as a required input for initial ritual depth
+- ⊗ Claim or require plan-item `effort` at session start to choose ceremony depth — that wiring deadlocks (no plan items yet)
+- ~ Ceremony dial (#3214) SHOULD use a **two-stage** path: start every session at **rapid** depth, then escalate to full ceremony when the plan lands M/L or the task proves project-shaped; optional intake-time provisional S/M/L is a tiebreaker only
+- ! Headless / autonomous runs MUST apply provisional estimates and stage transitions **without** operator confirmation prompts
+
+Rules:
+
+- ! When present, `effort` MUST be one of `S`, `M`, `L`, `XL` (case-sensitive)
+- ? Omit `effort` when sizing is unknown; validation does not require the field
+- ! XL items MUST be broken into S/M/L sub-items (or re-estimated to S/M/L) before `task scope:activate` / `task vbrief:activate` — both paths fail closed while any nested item still has `effort: "XL"`
+- ~ Swarm allocation SHOULD read `effort` when sizing agents (XL blocks dispatch until breakdown; L may warrant a dedicated agent; S/M are safe single-agent)
+- ~ Refinement SHOULD prompt for an estimate when accepting proposed → pending so cost signal is present before activation
+
+```json
+{
+  "id": "auth-slice",
+  "title": "Wire OAuth callback",
+  "status": "pending",
+  "effort": "M"
+}
+```
+
 ### Hierarchical Items (v0.6)
 
 Specs with phases, subphases, and tasks express nesting via `PlanItem.items`:

package/verification/verification.md

@@ -4,7 +4,7 @@ Goal-backward verification of agent output — check outcomes, not checklists.
 
 Legend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY.
 
-**⚠️ See also**: [coding/testing.md](../coding/testing.md) | [verification/uat.md](./uat.md) | [core/glossary.md](../glossary.md)
+**⚠️ See also**: [coding/testing.md](../coding/testing.md) | [verification/uat.md](./uat.md) | [core/glossary.md](../glossary.md) | [patterns/goal-gate-determinism.md](../patterns/goal-gate-determinism.md) (#852 — rigid goals/gates; flexible execution path) | [docs/gate-integrity.md](../docs/gate-integrity.md) (#3156 — do not clear red by editing the gate)
 
 > Adapted from [GSD](https://github.com/gsd-build/get-shit-done) verification model.
 
@@ -115,3 +115,4 @@ Pick the **strongest tier reachable** for each task.
 - ⊗ Accepting stubs as complete
 - ⊗ Asking a human to check what the agent can verify itself
 - ⊗ Verifying process ("I did steps 1–5") instead of outcomes ("user can log in")
+- ⊗ Clearing a failing gate by editing the gate definition, verifier, reward, required check, coverage floor, or policy flag — fix the work under test, or change the gate via issue/PR + review ([gate-integrity.md](../docs/gate-integrity.md) #3156)