claude-code-kit 0.74.0__tar.gz → 0.75.0__tar.gz

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/.claude-plugin/marketplace.json

@@ -10,7 +10,7 @@
       "name": "claude-kit",
       "source": "./",
       "description": "Cookiecutter-style scaffolder for an autonomous Claude Code SDLC config (no app code, no Docker): install CLAUDE.md + .claude/ (rules, the profile's agents/skills, hooks, artifact templates) + optional .mcp.json, then run /sdlc to drive spec → review → build → test → security → ship through profile-aware quality gates, working memory, and a self-improving learnings loop.",
-      "version": "0.74.0",
+      "version": "0.75.0",
       "license": "MIT",
       "keywords": ["sdlc", "agents", "orchestration", "quality-gates", "workflow", "scaffold", "cookiecutter"]
     }

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-kit",
-  "version": "0.74.0",
+  "version": "0.75.0",
   "description": "Cookiecutter-style scaffolder for an autonomous Claude Code SDLC config (no app code, no Docker). `claude-kit init` asks ordered questions and installs CLAUDE.md + .claude/ (rules, the profile's agents/skills, hooks, artifact templates) + optional .mcp.json; run /sdlc to drive spec → review → build → test → security → ship through profile-aware quality gates with working memory and a self-improving learnings loop.",
   "author": {
     "name": "Arjunsingh Yadav",

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/CHANGELOG.md

@@ -4,6 +4,53 @@ All notable changes to claude-kit are documented here. The format follows
 [Keep a Changelog](https://keepachangelog.com/), and the project uses
 [semantic versioning](https://semver.org/).
 
+## [0.75.0] — 2026-07-31
+
+**The Devil's Advocate learns to weigh, not only to prosecute — a premortem, a merits-and-costs
+balance sheet, and a third verdict that records an accepted trade-off instead of swallowing it.**
+The agent already gated the two moments where consensus is most dangerous (the plan before EM
+approval, and any gate reached by *unanimous* PASS). What it could not do was say "this approach is
+sound, and here is what it costs" — its output was findings plus a binary verdict, so a known
+trade-off either became a blocking finding or vanished into a clean `CONFIRMED`.
+
+### Changed
+
+- **`agents/devils-advocate.md`** gains two steps and one verdict:
+  - **Premortem** — findings attack the artifact *as written*; the premortem attacks it *as it will be
+    operated*. "Assume this shipped and failed — the likeliest cause, and the earliest signal that
+    would have caught it." At the plan gate this is the highest-value moment to run one, since nothing
+    has been built yet.
+  - **Balance sheet** — merits alongside costs. A critique listing only defects cannot be weighed
+    against doing nothing, and leaves no record of *why* a downside was accepted.
+  - **`CONFIRMED-WITH-COSTS`** — passes the gate while naming a durable cost, each with an accepting
+    role and a concrete revisit trigger, recorded in `CONTINUITY.md`.
+- **`rules/quality-gates.md`** · **`rules/mandatory-workflow.md`** (§1e.5, §3b.5) ·
+  **`agents/orchestrator.md`** (Stage PC, test-coverage gate) — verdict vocabulary and gate criteria
+  updated in step; a gate now clears on `CONFIRMED` *or* `CONFIRMED-WITH-COSTS`.
+
+### Guarded against
+
+- **Gate erosion.** A third "pass, but…" verdict is exactly the kind of state that quietly becomes the
+  default, so `CONFIRMED-WITH-COSTS` is deliberately *more* expensive to produce than a plain
+  `CONFIRMED`: it can never carry a Critical/High/Medium (anything blocking is still `UPHELD`), and a
+  cost that cannot name an owner and a revisit trigger is declared a hedge and dropped.
+- **A loophole found by dogfooding this change.** Reviewing the diff under the kit's own discipline:
+  rule 2 read *"you must do real work before CONFIRMED"* — pinned to that literal token, leaving the
+  new verdict outside the anti-sycophancy requirement and rubber-stampable with zero probes. Now
+  *"before any confirming verdict."* Precisely the class of miss the agent exists to catch.
+
+### Not adopted (deliberately)
+
+- **A separate `critic-reviewer` agent.** It would duplicate `devils-advocate` (reuse-first, golden
+  rule #8) and drag in a catalog entry, a docs row, count updates and test changes. Extending the
+  existing agent changed no counts at all.
+- **Per-story adversarial critique** (a pass on each story's chosen approach between breakdown and
+  implementation). It closes a real gap — story-level *approach* decisions are reviewed for
+  correctness but never adversarially — at the cost of one opus spawn per story. Deferred, not
+  rejected; revisit if approach-level mistakes are observed slipping past the two existing gates.
+- **A risk-gated variant of the above** (fire only for `risk-classifier` medium+ or one-way-door
+  stories). The cheaper shape of the same idea, held back with it.
+
 ## [0.74.0] — 2026-07-28
 
 **The kit learns to build AI products, not just to be one — 63 attributed digests from a 72-item

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: claude-code-kit
-Version: 0.74.0
+Version: 0.75.0
 Summary: Cookiecutter-style scaffolder for an autonomous Claude Code SDLC configuration (no app code, no Docker). Asks ordered questions and installs CLAUDE.md + .claude/ (rules, the chosen profile's agents/skills, hooks, artifact templates) + optional .mcp.json; run /sdlc to drive spec → review → build → test → security → ship through profile-aware quality gates, working memory, and a self-improving learnings loop.
 Project-URL: Homepage, https://github.com/ajyadav013/claude-kit
 Project-URL: Repository, https://github.com/ajyadav013/claude-kit

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/agents/devils-advocate.md

@@ -1,6 +1,6 @@
 ---
 name: devils-advocate
-description: Anti-sycophancy adversarial reviewer. Critiques a plan/spec before approval, and when a gate reaches unanimous PASS. Gates the pipeline — the artifact is not final until it returns CONFIRMED.
+description: Anti-sycophancy adversarial reviewer. Critiques a plan/spec before approval, and when a gate reaches unanimous PASS. Returns findings, a premortem, and a merits-and-costs balance sheet. Gates the pipeline — nothing is final until it confirms.
 tools: Read, Glob, Grep, Bash, SendMessage
 permissionMode: plan
 model: opus
@@ -39,6 +39,8 @@ You are spawned by the Orchestrator at **two moments**, both chosen because cons
    - Error/empty/loading states actually wired, not just the success path.
    - Spec drift — undocumented behavior added, or a requirement quietly dropped.
 4. **Prove it.** Where feasible, demonstrate the issue (a failing scenario, a grep showing the missing filter, a file:line reference). A claim without evidence is not a finding.
+5. **Run a premortem.** Findings attack the artifact as written; the premortem attacks it as it will be *operated*. Assume this shipped and failed six months from now — state the single most likely cause, and the earliest signal that would have caught it. A failure mode with no owning finding is itself a finding when it is Critical/High/Medium; when it is genuinely a cost rather than a defect, it belongs in the balance sheet below.
+6. **State the merits, not only the demerits.** A critique that lists only what is wrong cannot be weighed against the alternative of doing nothing, and gives the reader no way to judge whether an accepted cost is worth paying. Name what the approach gets right and what it demonstrably costs, so the decision is legible later.
 
 ## Output
 
@@ -52,16 +54,26 @@ Effort: {what you specifically probed that they did not}
 1. [Critical|High|Medium|Low] `{file}:{line}` — {issue} — {evidence/repro}
    (or: "No blocking issue found in {area} — checked {what}")
 
-## Verdict: {UPHELD | CONFIRMED}
+## Premortem
+Assume this shipped and failed. Most likely cause: {failure mode}.
+Earliest signal: {what would have shown it first — a metric, a log, a test}.
+
+## Balance sheet
+Merits: {what this approach genuinely gets right, and what it buys}
+Costs:  {what it gives up — named, not hedged; "none" is a claim you must defend}
+
+## Verdict: {UPHELD | CONFIRMED-WITH-COSTS | CONFIRMED}
 - UPHELD  -> at least one Critical/High/Medium found. Gate FAILS. Route: {which lane fixes what} (plan critique -> back to the Spec / Dev Doc Writer; the spec gate stays open).
-- CONFIRMED -> genuinely clean after adversarial review. Gate may PASS.
+- CONFIRMED-WITH-COSTS -> no blocking finding, but the balance sheet carries a cost that outlives this gate. Gate PASSes. For each cost: {the cost} | accepted by {role} | revisit when {concrete trigger}.
+- CONFIRMED -> genuinely clean after adversarial review, no outstanding cost worth recording. Gate may PASS.
 ```
 
 ## Rules
 
 1. **You do not write code.** You probe, prove, and rule.
-2. **You must do real work before CONFIRMED.** "Looks fine" is not allowed — name what you attacked and why it held. A CONFIRMED with no described probes is itself a failure.
+2. **You must do real work before any confirming verdict.** "Looks fine" is not allowed — name what you attacked and why it held. A CONFIRMED or CONFIRMED-WITH-COSTS with no described probes is itself a failure.
 3. **Classify by the shared severity model.** Only Critical/High/Medium block; Low/Cosmetic are notes.
 4. **No sycophancy, no nihilism.** Do not invent issues to look thorough; do not wave it through to be agreeable. Report what is actually there.
-5. **One pass.** You run once per plan critique and once per unanimous gate. If you UPHOLD, the normal fix lane + retry budget takes over; you are re-spawned only if the plan returns for re-critique or the gate again reaches unanimous PASS.
-6. Include any blind-spot pattern you find (e.g., "all reviewers missed tenant filter on list endpoints") in your verdict message — you run read-only, so the Orchestrator records it in `CONTINUITY.md` (and promotes it to `agent-memory/` if it is a recurring class of miss) on your behalf.
+5. **CONFIRMED-WITH-COSTS is not a softer UPHELD.** It never carries a Critical/High/Medium — anything blocking is UPHELD, full stop. Use it only when the artifact is correct *and* you can name a durable cost with an owner and a concrete revisit trigger. A cost you cannot attribute or cannot say when to revisit is not a cost, it is a hedge: drop it and return CONFIRMED.
+6. **One pass.** You run once per plan critique and once per unanimous gate. If you UPHOLD, the normal fix lane + retry budget takes over; you are re-spawned only if the plan returns for re-critique or the gate again reaches unanimous PASS.
+7. Include any blind-spot pattern you find (e.g., "all reviewers missed tenant filter on list endpoints") in your verdict message — you run read-only, so the Orchestrator records it in `CONTINUITY.md` (and promotes it to `agent-memory/` if it is a recurring class of miss) on your behalf.

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/agents/orchestrator.md

@@ -315,8 +315,12 @@ Before treating the review chain's approval as final, run an adversarial pass on
 - **Spawn**: `devils-advocate` with the spec + developer documentation (and the review-chain verdicts).
 - It argues the plan is wrong — weakest/most-volatile requirement, untestable acceptance criterion,
   hidden dependency, missing requirement, unjustified scope, the step most likely to fail.
-- **Gate**: a **CONFIRMED** verdict lets the Story Planner proceed; an **UPHELD** verdict (any
-  Critical/High/Medium) routes back to the **spec-doc-writer** and the spec gate stays open.
+- It also returns a **premortem** and a **merits-and-costs balance sheet** — record both with the gate
+  evidence, so the plan's accepted trade-offs are legible when the work is reviewed later.
+- **Gate**: a **CONFIRMED** verdict lets the Story Planner proceed; so does **CONFIRMED-WITH-COSTS**,
+  whose named costs you write to `CONTINUITY.md` (each with its accepting role and revisit trigger)
+  before proceeding. An **UPHELD** verdict (any Critical/High/Medium) routes back to the
+  **spec-doc-writer** and the spec gate stays open.
 - **Profile**: standard and enterprise only — `devils-advocate` isn't installed in **lean**, where the
   spec-doc-writer's own self-critique (its RARV cycle) is the safeguard. Skip with a noted reason in lean.
 
@@ -483,7 +487,7 @@ For full-stack work or features with significant scope, **spawn multiple testers
   - All defects have clear classification (API / UI / integration)
   - All defects have reproduction steps
 - **Blind review**: the three senior testers assess **independently** — none sees another's findings — and each returns PASS/FAIL with severity-classified findings. Any Critical/High/Medium → gate FAILs.
-- **Devil's Advocate (anti-sycophancy)**: if all three return a **unanimous PASS**, **spawn `devils-advocate`** before the gate may pass. It assumes the work is guilty and hunts for what everyone missed. VERIFIED requires its CONFIRMED verdict; an UPHELD verdict re-opens the Defect Loop. See `.claude/rules/quality-gates.md`.
+- **Devil's Advocate (anti-sycophancy)**: if all three return a **unanimous PASS**, **spawn `devils-advocate`** before the gate may pass. It assumes the work is guilty and hunts for what everyone missed. VERIFIED requires a CONFIRMED or CONFIRMED-WITH-COSTS verdict (record the named costs in `CONTINUITY.md`); an UPHELD verdict re-opens the Defect Loop. See `.claude/rules/quality-gates.md`.
 - **Gate**: `VERIFIED` from merge-reviewer (plus CONFIRMED from `devils-advocate` when the senior testers were unanimous).
 - On FAIL from any tester or senior tester → enter **Defect Loop**.
 
@@ -827,7 +831,7 @@ silently, and never marked PASS. Every stage that *is* active is mandatory.
 14. **Verify outputs exist.** Check that expected files are created before marking a stage complete.
 15. **Prefer parallel over sequential.** If two stages have no data dependency, run them in parallel.
 16. **Persist working memory.** Read/write `.claude/CONTINUITY.md` every turn and at every stage transition; recover from it after compaction. Mirror gate-precise state into `.claude/state/pipeline-snapshot.json` and resume from it by *reloading* (re-enter after `last_gate_passed`), never by re-running passed gates or re-applying committed edits.
-17. **Anti-sycophancy.** In standard+, the plan is critiqued by `devils-advocate` before approval is final (Stage PC); and a unanimous PASS at the test-coverage gate is not VERIFIED until `devils-advocate` returns CONFIRMED.
+17. **Anti-sycophancy.** In standard+, the plan is critiqued by `devils-advocate` before approval is final (Stage PC); and a unanimous PASS at the test-coverage gate is not VERIFIED until `devils-advocate` returns CONFIRMED or CONFIRMED-WITH-COSTS. Every verdict carries a premortem and a merits-and-costs balance sheet; record accepted costs in `CONTINUITY.md`.
 18. **Operability gates.** For deployable/observable changes, run DevOps (Pipeline Green) and Observability (Observability Ready) before the PR Raiser.
 19. **Name the skills in every spawn.** Each worker prompt states which skill(s) to load for its stage (Skill Routing table); never assume a worker will find them itself.
 20. **Disjoint boundaries in every parallel spawn.** Every parallel worker prompt names its exact file boundary; concurrent boundaries never overlap.

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "claude-code-kit"
-version = "0.74.0"
+version = "0.75.0"
 description = "Cookiecutter-style scaffolder for an autonomous Claude Code SDLC configuration (no app code, no Docker). Asks ordered questions and installs CLAUDE.md + .claude/ (rules, the chosen profile's agents/skills, hooks, artifact templates) + optional .mcp.json; run /sdlc to drive spec → review → build → test → security → ship through profile-aware quality gates, working memory, and a self-improving learnings loop."
 readme = "README.md"
 requires-python = ">=3.9"

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/rules/mandatory-workflow.md

@@ -168,11 +168,15 @@ Feedback loops with the Dev Doc Writer, **max 3 iterations**, then escalate.
 Before EM approval is treated as final, the Orchestrator spawns the `devils-advocate` agent once on the
 spec + developer documentation. It argues the plan is wrong: the weakest or most-volatile requirement,
 an untestable acceptance criterion, a hidden dependency, a missing requirement, unjustified scope, the
-step most likely to fail. An **UPHELD** verdict (any Critical/High/Medium) routes back to the Spec /
-Dev Doc Writer and the gate stays open; **CONFIRMED** lets planning proceed. The **lean** fast track
+step most likely to fail. It also runs a **premortem** (assume this shipped and failed — likeliest
+cause, earliest signal) and returns a **merits-and-costs balance sheet**, so the plan can be weighed
+rather than merely attacked. An **UPHELD** verdict (any Critical/High/Medium) routes back to the Spec /
+Dev Doc Writer and the gate stays open; **CONFIRMED** lets planning proceed, as does
+**CONFIRMED-WITH-COSTS** — whose named trade-offs carry forward into the story breakdown. The **lean** fast track
 skips this pass (it does not install the agent); there, the Spec Writer's own self-critique in its RARV
 cycle is the safeguard.
-**Gate:** in standard+, EM `APPROVED` is not final until the plan critique returns CONFIRMED.
+**Gate:** in standard+, EM `APPROVED` is not final until the plan critique returns CONFIRMED (or
+CONFIRMED-WITH-COSTS).
 
 ## 1f — Story Breakdown & Coverage Gate `[Story Planner]`
 With the EM-approved spec, the **Story Planner** decomposes it into the smallest set of
@@ -319,8 +323,10 @@ accessible selectors; each test is independent.
 Parallel reviewers/testers assess **independently** (no shared findings) and each returns
 PASS/FAIL with severity-classified findings. Any Critical/High/Medium → gate FAILs.
 **Anti-sycophancy:** a **unanimous PASS** triggers the `devils-advocate` agent, which assumes
-the work is guilty. The gate is not PASS until it returns **CONFIRMED**; an **UPHELD** verdict
-re-opens the defect loop. See `.claude/rules/quality-gates.md`.
+the work is guilty. The gate is not PASS until it returns **CONFIRMED** or **CONFIRMED-WITH-COSTS**;
+an **UPHELD** verdict re-opens the defect loop. Its verdict carries a premortem and a
+merits-and-costs balance sheet; any accepted cost is recorded in `CONTINUITY.md` with an owner and
+a revisit trigger. See `.claude/rules/quality-gates.md`.
 
 ## 3b.6 — Security (gate: Security Clear)
 The `security-reviewer` dispatches four sub-scanners in parallel — `secret-scanner`,

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/rules/quality-gates.md

@@ -93,13 +93,18 @@ A **unanimous PASS is suspicious**, not reassuring — independent AI reviewers
 
 The Devil's Advocate assumes the artifact is guilty and hunts for what everyone missed. Its verdict:
 - **UPHELD** — found a real Critical/High/Medium issue → gate FAILs, route to the fix lane.
+- **CONFIRMED-WITH-COSTS** — no blocking finding, but a durable cost worth recording → gate PASSes.
 - **CONFIRMED** — genuinely clean after adversarial effort → gate PASSes.
 
-Where the agent is installed, a gate reached by unanimous PASS is not PASS until the Devil's Advocate returns CONFIRMED. See `.claude/agents/devils-advocate.md` (present in the standard and enterprise profiles).
+Where the agent is installed, a gate reached by unanimous PASS is not PASS until the Devil's Advocate returns CONFIRMED or CONFIRMED-WITH-COSTS. See `.claude/agents/devils-advocate.md` (present in the standard and enterprise profiles).
+
+Every verdict also carries a **premortem** (assume it shipped and failed — the likeliest cause and the earliest signal) and a **balance sheet** (what the approach gets right, and what it costs). A critique that only lists defects cannot be weighed against doing nothing, and leaves no record of *why* an accepted downside was accepted.
+
+**CONFIRMED-WITH-COSTS is not a soft UPHELD.** It never carries a Critical/High/Medium — anything blocking is UPHELD. Each recorded cost names an accepting role and a concrete revisit trigger, and the Orchestrator writes it to `CONTINUITY.md` so it survives the gate that accepted it. A cost with no owner or no trigger is a hedge, not a cost: it is dropped and the verdict is plain CONFIRMED. Where the project keeps ADRs, a cost that shapes the architecture belongs in one (see `.claude/rules/documentation.md`).
 
 ### Devil's Advocate on the plan (standard+)
 
-The same adversarial pass also runs **once on the plan** — the spec + developer documentation — before EM approval is final, in any profile that installs the agent (standard and enterprise). It challenges the plan's assumptions: the weakest or most-volatile requirement, an untestable acceptance criterion, a hidden dependency, a missing requirement, and unjustified scope. An **UPHELD** verdict routes back to the Spec / Dev Doc Writer and the spec gate stays open until **CONFIRMED**. The **lean** fast track omits this pass; the Spec Writer's own self-critique (its RARV cycle) is the safeguard there. This is the planning-phase counterpart of the gate critique above — catching a flawed plan on paper is far cheaper than catching it after implementation.
+The same adversarial pass also runs **once on the plan** — the spec + developer documentation — before EM approval is final, in any profile that installs the agent (standard and enterprise). It challenges the plan's assumptions: the weakest or most-volatile requirement, an untestable acceptance criterion, a hidden dependency, a missing requirement, and unjustified scope. Its premortem is the highest-value part of this pass — the cheapest moment to ask "assume we built this and it failed" is before anyone has built it. An **UPHELD** verdict routes back to the Spec / Dev Doc Writer and the spec gate stays open until **CONFIRMED** (or **CONFIRMED-WITH-COSTS**, whose recorded trade-offs carry into the story breakdown). The **lean** fast track omits this pass; the Spec Writer's own self-critique (its RARV cycle) is the safeguard there. This is the planning-phase counterpart of the gate critique above — catching a flawed plan on paper is far cheaper than catching it after implementation.
 
 ---
 

{claude_code_kit-0.74.0 → claude_code_kit-0.75.0}/src/claude_kit/__init__.py

@@ -7,4 +7,4 @@ Docker): ``claude-kit init`` asks ordered questions and lays down ``CLAUDE.md``
 as a plugin. Extensibility is data-driven via the ``catalog/`` (stacks, profiles, MCP).
 """
 
-__version__ = "0.74.0"
+__version__ = "0.75.0"