@warnyin/sdlc 0.3.0 → 0.5.0

package/CHANGELOG.md

@@ -1,95 +1,147 @@
-# Changelog
-
-## 0.3.0 (2026-08-21)
-
-- **`/sdlc:auto` resumes an open change** instead of always starting at `new`. It
-  resolves its entry stage from `sdlc status` first: an argument naming an active
-  change maps that change's status to the entry stage and the pipeline starts
-  there, keeping the tier, Delta and Assumptions it was already triaged with. Only
-  an argument matching no active change starts at `new`. The entry stage is stated
-  in the plan line, so a resume is never silent. The status → stage table stays in
-  `next.md` alone rather than being copied into a second place that can drift.
-- **Fix (ownership)**: `installFile` dropped a file's manifest entry whenever it
-  kept the file. The next run then saw a path it had never installed, which
-  permanently disarmed `update`'s refresh branch — the file froze at its old
-  payload version and every later run relabelled it user-modified. That affects
-  anyone who re-runs `init` to upgrade before `update`. A kept file now carries
-  its recorded hash forward; prune is unaffected (its guard compares the file on
-  disk against that same hash) and in fact strictly safer, since a kept file is no
-  longer even a prune candidate.
-- A file whose content still matches its recorded hash is reported as
-  `kept (ours, older version — run update to refresh)` instead of
-  `kept (user-modified)`, which sent people hunting for an edit they never made.
-
-## 0.2.2 (2026-08-20)
-
-- `init` (and `update`) now drop a `.gitkeep` in `sdlc/changes/archive/`, so the
-  directory survives a commit and is still there after a clone. 0.2.1 made
-  `archive` recover from the missing directory; this stops it going missing.
-  The marker is not payload-owned — prune never reclaims it and the installer
-  never warns about it.
-
-## 0.2.1 (2026-08-20)
-
-- **Fix**: `archive` failed with `ENOENT` on the first change a repo ever ships.
-  `init` scaffolds `sdlc/changes/archive/`, but git does not track empty
-  directories, so the folder is absent for everyone who clones before that first
-  ship. The rename is now preceded by a `mkdir -p` of the archive root.
-
-  The failure landed mid-phase-2, after the delta had been merged into the living
-  specs, evals promoted, `status: shipped` stamped and the ship event journalled —
-  a repo left half-shipped while the CLI reported total failure. The directory is
-  now prepared next to the other destination checks, before phase 1 computes a
-  single merge, so an unusable archive path aborts with the specs untouched.
-
-## 0.2.0 (2026-08-20)
-
-`init` is now an installer you can actually see working — still zero dependencies.
-
-- **Interactive tool picker**: a searchable checkbox list replaces the
-  comma-separated typing prompt. Arrows move, `space` toggles, typing filters,
-  `ctrl+a` selects everything on screen, `enter` confirms, `ctrl+c` cancels
-  without installing anything.
-- **Tool detection**: tools the project already uses (`.claude/`, `.cursor/`,
-  `AGENTS.md`, …) come pre-selected; an empty project still defaults to claude.
-- **Post-install summary**: artifact counts, the adapter path per tool, a
-  written/unchanged/refreshed/kept tally, and Getting-started hints that differ
-  per tool (slash commands for Claude Code, prose for the rest).
-- **Colour** gated on `NO_COLOR` > `FORCE_COLOR` > TTY, with an ASCII glyph
-  fallback for legacy Windows consoles.
-- `--tool` accepts `all` and `none`; `--tools` is an alias. An empty `--tool`,
-  an unknown tool, or `all` mixed with a named tool now fails loudly.
-- **Fix**: `update` read an explicit `tools: []` (what `init --tool none` writes)
-  as "unset" and reinstalled claude. A missing key and a declared-empty one are
-  now distinguished.
-
-## 0.1.2 (2026-08-20)
-
-- Fix: `readStdinJson()` grew a 1s grace timeout — journal and hook utilities no
-  longer hang when stdin is open but idle.
-- Fix: hooks release stdin (pause + unref) so an open-idle stdin cannot keep a
-  hook process alive.
-
-## 0.1.1 (2026-08-20)
-
-- Fix: the entrypoint guard must realpath `process.argv[1]` — npx invokes through
-  a `node_modules/.bin` symlink, which made the CLI a silent no-op.
-- Fix (CI): `node --test` bare discovery; a quoted glob is not expanded on Node 20.
-
-## 0.1.0 (2026-08-20)
-
-Initial release — the full Day-1 SDLC loop:
-
-- CLI: `init` (multi-tool: claude/cursor/windsurf/copilot/cline/gemini/agents-md),
-  `update` (ownership-aware refresh + guarded prune), `validate`, `status`,
-  `observe`, `archive`.
-- Artifact model: constitution (≤30) + steering with inclusion modes + harness
-  (routing/triage/autonomy policy) + living specs + delta-based changes
-  (vibe/standard/deep caps 40/100/150) + contracts (tests ≤60, evals ≤40).
-- Managed Claude Code hooks: static-context injector, spec write-lock with TTL
-  gates, artifact validator + steering pointers, session token/cost journaling,
-  compact-event tracking.
-- 13 stage playbooks + 13 `/sdlc:*` commands + 3 background skills + 8 agents
-  with model routing (cheap/balanced/deepest).
-- Observability: per-change tokens/cost, first-pass rate, lead time, dead
-  steering and residency flags; post-ship learner loop (distill, never bloat).
+# Changelog
+
+## 0.5.0 (2026-08-25)
+
+- **`--auto` on every pipeline stage.** `/sdlc:auto` already ran the whole pipeline,
+  but it stopped at every escalation — so you were pulled back in three or four times
+  per change and typed each stage anyway. Now all seven stage commands take `--auto`:
+  the run gathers what it needs, confirms once, and goes to ship. The confirmation is
+  decidable item by item — scope as understood, the tier and why, every ambiguity with
+  the assumption to be acted on, and each escalation as its own refusable line, the
+  ship row naming the hard-floor surface it covers instead of hiding behind a general
+  "run without me". Nothing is written before you confirm, down to the active-change
+  pointer, so declining leaves the repository untouched. The approval covers that run
+  only — not config, not the next change, not a resume. Anything outside what you
+  confirmed still stops and asks.
+- Escalations passed unattended are journalled, counted by `/sdlc:observe` as
+  `unattended×N`, and listed in the digest: the record shows where a human would
+  normally have stood and, that run, did not.
+- **Verify and review outcomes now record how they were produced** (`mode=panel|solo`).
+  A journal that says "verify passed" hides the thing a reader most needs later —
+  whether that verdict came from independent reviewers or from the same loop that
+  wrote the code. `observe` marks such changes `self-judged`, and the digest must name
+  self-produced outcomes. Absent provenance reads as unknown, never as `panel`, so
+  older journals are not retroactively dressed up as independently reviewed; where
+  provenance is mixed, the weakest link decides.
+- Where a panel cannot run, the playbooks now say to judge in the main loop and record
+  that — not to skip the stage. A review that never happened is worse than one
+  labelled honestly.
+- The constitution gains a hard rule: human-written text SHALL NOT reach a shell as an
+  argument. It is the defect that got past two separate gates in 0.4.0.
+
+## 0.4.0 (2026-08-25)
+
+- **New stage command `/sdlc:feedback`** — reports a bug, a rough edge, or a missing
+  feature in the framework itself to `warnyin/warnyin-sdlc`, from inside the session
+  where you hit it. It collects the context a maintainer triages by (framework and
+  Node version, OS, tool adapter, active change id and status), redacts it, shows you
+  the complete draft, and files it only after you approve. Submission goes through
+  `gh`; when `gh` is missing, logged out, or authenticated only against an enterprise
+  host, you get a prefilled issue URL instead — a normal path, not an error.
+  Nothing is attached automatically: no logs, no journal, no diff. Redaction is a
+  rule list rather than a guarantee, and the playbook says so — your eyes on the
+  draft are the control.
+- Human-written text never reaches a shell as an argument: the body travels over
+  stdin, while the title and the duplicate-search keywords are written by the agent
+  under a length and character allow-list instead of being pasted raw.
+- **New: `warnyin-sdlc version`** (also `--version` / `-v`). Nothing in an installed
+  project was readable as a version — an npx install leaves no package behind — so
+  every bug report would have carried `unknown` in the field that decides whether a
+  report can be acted on at all.
+- The repository now ships `.github/ISSUE_TEMPLATE/` bug and feature forms asking for
+  the same fields the command collects, so web-filed and command-filed reports read
+  alike.
+
+## 0.3.0 (2026-08-21)
+
+- **`/sdlc:auto` resumes an open change** instead of always starting at `new`. It
+  resolves its entry stage from `sdlc status` first: an argument naming an active
+  change maps that change's status to the entry stage and the pipeline starts
+  there, keeping the tier, Delta and Assumptions it was already triaged with. Only
+  an argument matching no active change starts at `new`. The entry stage is stated
+  in the plan line, so a resume is never silent. The status → stage table stays in
+  `next.md` alone rather than being copied into a second place that can drift.
+- **Fix (ownership)**: `installFile` dropped a file's manifest entry whenever it
+  kept the file. The next run then saw a path it had never installed, which
+  permanently disarmed `update`'s refresh branch — the file froze at its old
+  payload version and every later run relabelled it user-modified. That affects
+  anyone who re-runs `init` to upgrade before `update`. A kept file now carries
+  its recorded hash forward; prune is unaffected (its guard compares the file on
+  disk against that same hash) and in fact strictly safer, since a kept file is no
+  longer even a prune candidate.
+- A file whose content still matches its recorded hash is reported as
+  `kept (ours, older version — run update to refresh)` instead of
+  `kept (user-modified)`, which sent people hunting for an edit they never made.
+
+## 0.2.2 (2026-08-20)
+
+- `init` (and `update`) now drop a `.gitkeep` in `sdlc/changes/archive/`, so the
+  directory survives a commit and is still there after a clone. 0.2.1 made
+  `archive` recover from the missing directory; this stops it going missing.
+  The marker is not payload-owned — prune never reclaims it and the installer
+  never warns about it.
+
+## 0.2.1 (2026-08-20)
+
+- **Fix**: `archive` failed with `ENOENT` on the first change a repo ever ships.
+  `init` scaffolds `sdlc/changes/archive/`, but git does not track empty
+  directories, so the folder is absent for everyone who clones before that first
+  ship. The rename is now preceded by a `mkdir -p` of the archive root.
+
+  The failure landed mid-phase-2, after the delta had been merged into the living
+  specs, evals promoted, `status: shipped` stamped and the ship event journalled —
+  a repo left half-shipped while the CLI reported total failure. The directory is
+  now prepared next to the other destination checks, before phase 1 computes a
+  single merge, so an unusable archive path aborts with the specs untouched.
+
+## 0.2.0 (2026-08-20)
+
+`init` is now an installer you can actually see working — still zero dependencies.
+
+- **Interactive tool picker**: a searchable checkbox list replaces the
+  comma-separated typing prompt. Arrows move, `space` toggles, typing filters,
+  `ctrl+a` selects everything on screen, `enter` confirms, `ctrl+c` cancels
+  without installing anything.
+- **Tool detection**: tools the project already uses (`.claude/`, `.cursor/`,
+  `AGENTS.md`, …) come pre-selected; an empty project still defaults to claude.
+- **Post-install summary**: artifact counts, the adapter path per tool, a
+  written/unchanged/refreshed/kept tally, and Getting-started hints that differ
+  per tool (slash commands for Claude Code, prose for the rest).
+- **Colour** gated on `NO_COLOR` > `FORCE_COLOR` > TTY, with an ASCII glyph
+  fallback for legacy Windows consoles.
+- `--tool` accepts `all` and `none`; `--tools` is an alias. An empty `--tool`,
+  an unknown tool, or `all` mixed with a named tool now fails loudly.
+- **Fix**: `update` read an explicit `tools: []` (what `init --tool none` writes)
+  as "unset" and reinstalled claude. A missing key and a declared-empty one are
+  now distinguished.
+
+## 0.1.2 (2026-08-20)
+
+- Fix: `readStdinJson()` grew a 1s grace timeout — journal and hook utilities no
+  longer hang when stdin is open but idle.
+- Fix: hooks release stdin (pause + unref) so an open-idle stdin cannot keep a
+  hook process alive.
+
+## 0.1.1 (2026-08-20)
+
+- Fix: the entrypoint guard must realpath `process.argv[1]` — npx invokes through
+  a `node_modules/.bin` symlink, which made the CLI a silent no-op.
+- Fix (CI): `node --test` bare discovery; a quoted glob is not expanded on Node 20.
+
+## 0.1.0 (2026-08-20)
+
+Initial release — the full Day-1 SDLC loop:
+
+- CLI: `init` (multi-tool: claude/cursor/windsurf/copilot/cline/gemini/agents-md),
+  `update` (ownership-aware refresh + guarded prune), `validate`, `status`,
+  `observe`, `archive`.
+- Artifact model: constitution (≤30) + steering with inclusion modes + harness
+  (routing/triage/autonomy policy) + living specs + delta-based changes
+  (vibe/standard/deep caps 40/100/150) + contracts (tests ≤60, evals ≤40).
+- Managed Claude Code hooks: static-context injector, spec write-lock with TTL
+  gates, artifact validator + steering pointers, session token/cost journaling,
+  compact-event tracking.
+- 13 stage playbooks + 13 `/sdlc:*` commands + 3 background skills + 8 agents
+  with model routing (cheap/balanced/deepest).
+- Observability: per-change tokens/cost, first-pass rate, lead time, dead
+  steering and residency flags; post-ship learner loop (distill, never bloat).

package/README.md

@@ -1,84 +1,91 @@
-# @warnyin/sdlc
-
-**Spec-driven, AI-driven SDLC for coding agents — token-lean by construction.**
-
-Operationalizes the *"New SDLC with Vibe Coding"* (Day-1) work process: the human configures
-the harness once; the AI drives changes end-to-end through contract-first gates; deterministic
-hooks and a validator enforce the rules; a journal prices every change in real tokens.
-
-Inspired by OpenSpec (delta specs, archive lifecycle), spec-kit (artifact grammar), and
-Kiro (steering + enforced hooks) — tuned for minimum context residency.
-
-## Install
-
-```bash
-cd your-project
-npx @warnyin/sdlc init          # interactive picker; tools already in the project are pre-selected
-```
-
-The picker is a checkbox list — arrows move, `space` toggles, typing filters, `ctrl+a` selects
-everything on screen, `enter` confirms. Skip it in CI or scripts:
-
-```bash
-npx @warnyin/sdlc init --tool claude,cursor   # explicit list
-npx @warnyin/sdlc init --tool all             # every supported tool
-npx @warnyin/sdlc init --tool none            # sdlc/ framework only, no agent adapters
-```
-
-Then in your coding agent:
-
-```
-/sdlc:init                      # interview → constitution + harness (the one human gate)
-/sdlc:auto Add rate limiting    # AI runs new → contract → build → verify → ship
-/sdlc:auto add-rate-limiting    # already opened it with /sdlc:new? auto resumes from there
-```
-
-## How it works
-
-```
-sdlc/
-├── context/constitution.md   ≤30 lines — the ONLY always-loaded prose (hook-injected)
-├── context/steering/*.md     scoped knowledge · inclusion: always|paths|manual|agent
-├── harness.md                tools, model routing, tier triage, autonomy policy
-├── specs/<capability>/       living specs — WHEN/THEN SHALL, merged mechanically at ship
-├── changes/<id>/             one change: change.md (Why+Delta+Tasks) + contract/ + journal
-└── changes/archive/          shipped changes + digests (the async human touchpoint)
-```
-
-- **Contract-first**: tests + evals are written *before* code and gate everything after.
-- **AI-driven, exception-only humans**: the autonomy policy in `harness.md` decides what
-  auto-ships and what escalates (hard-floor: security/payments/data-loss/irreversible).
-- **Managed hooks** (Claude Code): SessionStart injects ≤60 lines of static context;
-  PreToolUse write-locks living specs; PostToolUse validates caps and points at steering;
-  Stop journals real token usage. Other tools get the same rules as prose + the validator.
-- **Self-improving, leaner over time**: a post-ship learner proposes rules with evidence;
-  the always-loaded budget is fixed, so learning must distill, not accumulate.
-- **Measured**: `npx @warnyin/sdlc observe` — tokens/cost per change, first-pass rate,
-  lead time, dead steering, context-overflow flags.
-
-## CLI
-
-```
-warnyin-sdlc init [--tool all|none|a,b]   scaffold + adapters + hooks (picker when omitted)
-warnyin-sdlc update [--force]       refresh payload, guarded prune of stale files
-warnyin-sdlc validate [id] [--strict]
-warnyin-sdlc status | observe [--json]
-warnyin-sdlc archive <id>           merge deltas into living specs + archive
-```
-
-## Commands (in your agent)
-
-`/sdlc:init` · `/sdlc:auto` · `/sdlc:new` · `/sdlc:design` · `/sdlc:contract` · `/sdlc:build`
-· `/sdlc:verify` · `/sdlc:review` · `/sdlc:ship` · `/sdlc:observe` · `/sdlc:converge`
-· `/sdlc:steer` · `/sdlc:next`
-
-Playbooks live in `sdlc/.playbook/` — behavior is defined once there; commands are thin pointers.
-
-## Development
-
-Zero dependencies, Node ≥ 20. `npm test` runs the black-box suite (temp dirs, real CLI spawns).
-This repo self-hosts: its own development flows through `sdlc/changes/`. After cloning, run
-`npm run setup:dogfood` to regenerate the installer-owned mirrors (`sdlc/.playbook/`,
-`sdlc/.hooks/`, `.claude/`).
-
-MIT
+# @warnyin/sdlc
+
+**Spec-driven, AI-driven SDLC for coding agents — token-lean by construction.**
+
+Operationalizes the *"New SDLC with Vibe Coding"* (Day-1) work process: the human configures
+the harness once; the AI drives changes end-to-end through contract-first gates; deterministic
+hooks and a validator enforce the rules; a journal prices every change in real tokens.
+
+Inspired by OpenSpec (delta specs, archive lifecycle), spec-kit (artifact grammar), and
+Kiro (steering + enforced hooks) — tuned for minimum context residency.
+
+## Install
+
+```bash
+cd your-project
+npx @warnyin/sdlc init          # interactive picker; tools already in the project are pre-selected
+```
+
+The picker is a checkbox list — arrows move, `space` toggles, typing filters, `ctrl+a` selects
+everything on screen, `enter` confirms. Skip it in CI or scripts:
+
+```bash
+npx @warnyin/sdlc init --tool claude,cursor   # explicit list
+npx @warnyin/sdlc init --tool all             # every supported tool
+npx @warnyin/sdlc init --tool none            # sdlc/ framework only, no agent adapters
+```
+
+Then in your coding agent:
+
+```
+/sdlc:init                      # interview → constitution + harness (the one human gate)
+/sdlc:auto Add rate limiting    # AI runs new → contract → build → verify → ship
+/sdlc:auto add-rate-limiting    # already opened it with /sdlc:new? auto resumes from there
+/sdlc:new Add rate limiting --auto   # any stage takes --auto: confirm once, then run to ship
+```
+
+`--auto` asks everything up front — scope, tier, each ambiguity, and every escalation
+it wants pre-approved as its own line you can refuse — then runs unattended. Nothing
+is written until you confirm, the approval covers that run only, and anything you did
+not pre-approve still stops and asks.
+
+## How it works
+
+```
+sdlc/
+├── context/constitution.md   ≤30 lines — the ONLY always-loaded prose (hook-injected)
+├── context/steering/*.md     scoped knowledge · inclusion: always|paths|manual|agent
+├── harness.md                tools, model routing, tier triage, autonomy policy
+├── specs/<capability>/       living specs — WHEN/THEN SHALL, merged mechanically at ship
+├── changes/<id>/             one change: change.md (Why+Delta+Tasks) + contract/ + journal
+└── changes/archive/          shipped changes + digests (the async human touchpoint)
+```
+
+- **Contract-first**: tests + evals are written *before* code and gate everything after.
+- **AI-driven, exception-only humans**: the autonomy policy in `harness.md` decides what
+  auto-ships and what escalates (hard-floor: security/payments/data-loss/irreversible).
+- **Managed hooks** (Claude Code): SessionStart injects ≤60 lines of static context;
+  PreToolUse write-locks living specs; PostToolUse validates caps and points at steering;
+  Stop journals real token usage. Other tools get the same rules as prose + the validator.
+- **Self-improving, leaner over time**: a post-ship learner proposes rules with evidence;
+  the always-loaded budget is fixed, so learning must distill, not accumulate.
+- **Measured**: `npx @warnyin/sdlc observe` — tokens/cost per change, first-pass rate,
+  lead time, dead steering, context-overflow flags.
+
+## CLI
+
+```
+warnyin-sdlc init [--tool all|none|a,b]   scaffold + adapters + hooks (picker when omitted)
+warnyin-sdlc update [--force]       refresh payload, guarded prune of stale files
+warnyin-sdlc validate [id] [--strict]
+warnyin-sdlc status | observe [--json]
+warnyin-sdlc archive <id>           merge deltas into living specs + archive
+warnyin-sdlc version | --version    print the installed framework version
+```
+
+## Commands (in your agent)
+
+`/sdlc:init` · `/sdlc:auto` · `/sdlc:new` · `/sdlc:design` · `/sdlc:contract` · `/sdlc:build`
+· `/sdlc:verify` · `/sdlc:review` · `/sdlc:ship` · `/sdlc:observe` · `/sdlc:converge`
+· `/sdlc:steer` · `/sdlc:next` · `/sdlc:feedback`
+
+Playbooks live in `sdlc/.playbook/` — behavior is defined once there; commands are thin pointers.
+
+## Development
+
+Zero dependencies, Node ≥ 20. `npm test` runs the black-box suite (temp dirs, real CLI spawns).
+This repo self-hosts: its own development flows through `sdlc/changes/`. After cloning, run
+`npm run setup:dogfood` to regenerate the installer-owned mirrors (`sdlc/.playbook/`,
+`sdlc/.hooks/`, `.claude/`).
+
+MIT

package/bin/cli.mjs

@@ -45,7 +45,7 @@ export function sha256(content) {
 }
 
 export function parseArgs(argv) {
-  const args = { _: [], tool: null, toolProvided: false, strict: false, force: false, json: false, help: false };
+  const args = { _: [], tool: null, toolProvided: false, strict: false, force: false, json: false, help: false, version: false };
   for (let i = 0; i < argv.length; i++) {
     const a = argv[i];
     if (a === '--tool' || a === '--tools') {
@@ -56,6 +56,7 @@ export function parseArgs(argv) {
     else if (a === '--force') args.force = true;
     else if (a === '--json') args.json = true;
     else if (a === '--help' || a === '-h') args.help = true;
+    else if (a === '--version' || a === '-v') args.version = true;
     else if (a.startsWith('--')) console.warn(`unknown flag ${a} (ignored)`);
     else args._.push(a);
   }
@@ -526,6 +527,12 @@ function runValidate(projectRoot, args) {
   process.exitCode = res.status ?? 0;
 }
 
+// Read from our own package.json: an npx install leaves nothing readable in the
+// target project, and a report whose version is `unknown` cannot be triaged.
+function pkgVersion() {
+  return JSON.parse(fs.readFileSync(path.join(PKG_ROOT, 'package.json'), 'utf8')).version;
+}
+
 const HELP = `@warnyin/sdlc — spec-driven AI-SDLC framework
 
 usage: warnyin-sdlc <command> [options]
@@ -536,14 +543,18 @@ usage: warnyin-sdlc <command> [options]
   status [--json]                   list active changes and their stage
   observe [--json]                  tokens/cost per change, residency, steering hits, drift flags
   archive <id>                      merge delta specs into living specs and archive the change
+  version | --version | -v          print the installed framework version
   help                              this text
 `;
 
 export async function main(argv = process.argv.slice(2), projectRoot = process.cwd()) {
   const args = parseArgs(argv);
   const cmd = args._[0];
-  if (args.help || !cmd || cmd === 'help') { console.log(HELP); return; }
   try {
+    // before the help branch: `--version` carries no command, and `!cmd` would
+    // otherwise print help instead of the version.
+    if (args.version || cmd === 'version') { console.log(pkgVersion()); return; }
+    if (args.help || !cmd || cmd === 'help') { console.log(HELP); return; }
     if (cmd === 'init') await cmdInit(projectRoot, args);
     else if (cmd === 'update') cmdUpdate(projectRoot, args);
     else if (cmd === 'validate') runValidate(projectRoot, args);