godpowers 1.6.10 → 1.6.12

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -5,6 +5,76 @@ All notable changes to Godpowers will be documented in this file.
5
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
6
6
  and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
7
7
 
8
+ ## [1.6.12] - 2026-05-16
9
+
10
+ Executable dashboard CLI and shared runtime status engine.
11
+
12
+ ### Added
13
+ - Added `lib/dashboard.js`, a shared executable engine for project status,
14
+ progress percentage, tier position, planning visibility, proactive checks,
15
+ open items, and recommended next command.
16
+ - Added `godpowers status --project .` for human, CI, and host-runtime status
17
+ checks outside the slash command layer.
18
+ - Added `godpowers next --project .` for direct next-action routing from disk
19
+ state.
20
+ - Added `--json` output for the new status and next commands.
21
+ - Added dashboard behavioral tests, CLI status and next tests, and git
22
+ porcelain parsing tests.
23
+
24
+ ### Changed
25
+ - `/god-status`, `/god-next`, `/god-mode`, and `god-orchestrator` now point to
26
+ the shared dashboard engine when local runtime execution is available.
27
+ - The installer help now documents status and next as first-class commands.
28
+ - Release documentation now names the executable dashboard contract instead of
29
+ treating progress visibility as Markdown-only guidance.
30
+
31
+ ### Fixed
32
+ - Fixed git porcelain parsing so leading-space worktree entries such as
33
+ ` M README.md` do not clip filenames in dashboard output.
34
+
35
+ ### Validation
36
+ - Added tests prove dashboard computation, dashboard rendering, CLI JSON
37
+ output, proactive review suggestions, and staged path reporting.
38
+
39
+ ## [1.6.11] - 2026-05-16
40
+
41
+ Auto-invoke visibility and platform-neutral spawning patch.
42
+
43
+ ### Added
44
+ - Added a core auto-invoke visibility rule requiring Godpowers to show when it
45
+ automatically runs a command, spawns an agent, or calls a local runtime
46
+ helper.
47
+ - Added a proactive auto-invoke policy with four levels: read-only
48
+ suggestions, visible local helpers, bounded specialist agent spawns, and
49
+ explicit-approval-only actions.
50
+ - Added proactive checks to `/god-next`, `/god-status`, and God Mode closeouts
51
+ so users can see checkpoint, review, sync, docs, runtime, security,
52
+ dependency, and hygiene opportunities without asking.
53
+ - Added docs drift, runtime verification, and review queue guidance for
54
+ proactive closeouts.
55
+ - Added `docs/auto-invoke-visibility.md` as a local design note for the
56
+ auto-invoke policy.
57
+
58
+ ### Changed
59
+ - Replaced Claude-specific "Task tool" spawning wording in high-traffic skills
60
+ with platform-neutral host-agent spawning language.
61
+ - Clarified that Claude, Codex, Cursor, Windsurf, Gemini, OpenCode, Copilot,
62
+ Augment, Trae, Cline, Kilo, Antigravity, Qwen, CodeBuddy, and Pi may expose
63
+ specialist spawning differently while sharing the same Markdown agent
64
+ contracts.
65
+ - Updated the Codex installer path to replace skill directories before copying
66
+ `SKILL.md`, preventing stale nested files from older install shapes.
67
+ - `/god-sync`, `/god-scan`, `god-updater`, and `god-orchestrator` now describe
68
+ local sync helpers separately from spawned agents.
69
+
70
+ ### Guardrails
71
+ - Production launch, guessed staging URLs, provider dashboards, credentials,
72
+ destructive repair, review clearing, Critical security acceptance, git
73
+ staging, commits, pushes, packages, releases, and npm publishing still require
74
+ explicit user intent.
75
+ - If a host platform cannot provide a true fresh-context spawn, Godpowers must
76
+ say so instead of pretending a background agent ran.
77
+
8
78
  ## [1.6.10] - 2026-05-16
9
79
 
10
80
  Progress visibility and plain-language closeout patch.
package/README.md CHANGED
@@ -2,7 +2,7 @@
2
2
 
3
3
  [![CI](https://github.com/aihxp/godpowers/actions/workflows/ci.yml/badge.svg)](https://github.com/aihxp/godpowers/actions/workflows/ci.yml)
4
4
  [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
5
- [![Version](https://img.shields.io/badge/version-1.6.10-blue)](CHANGELOG.md)
5
+ [![Version](https://img.shields.io/badge/version-1.6.12-blue)](CHANGELOG.md)
6
6
  [![npm](https://img.shields.io/npm/v/godpowers.svg)](https://www.npmjs.com/package/godpowers)
7
7
 
8
8
  **Ship fast. Ship right. Ship everything. Ship accountably.**
@@ -12,10 +12,10 @@ idea to hardened production. It runs as **slash commands inside your AI coding
12
12
  tool** (Claude Code, Codex, Cursor, etc.) that orchestrate **specialist agents**
13
13
  in fresh contexts to do the work.
14
14
 
15
- Version 1.6.10 keeps the stable Godpowers surface while making progress more
16
- visible: user-facing output now prefers "project run" over internal "arc"
17
- jargon and status closeouts show PRD, roadmap, milestone, and completion
18
- position when available.
15
+ Version 1.6.12 adds an executable dashboard engine and CLI status surface:
16
+ `godpowers status --project .` and `godpowers next --project .` now compute the
17
+ same disk-derived progress, planning visibility, proactive checks, and next
18
+ action used by `/god-status`, `/god-next`, and God Mode closeouts.
19
19
 
20
20
  It fuses four disciplines into one unified workflow:
21
21
 
@@ -50,6 +50,13 @@ The installer copies:
50
50
  - Codex agent metadata to `<runtime>/agents/*.toml`
51
51
  - SessionStart hook (Claude Code only) to `<runtime>/hooks/`
52
52
 
53
+ Agent spawning is host-native. Claude uses its native agent/task interface,
54
+ Codex uses installed `agents/*.toml` metadata backed by the same Markdown agent
55
+ contracts, and the other runtimes use their supported agent or subagent
56
+ mechanism against the installed `agents/god-*.md` files. If a host cannot
57
+ provide a true fresh-context spawn, Godpowers must report that limitation
58
+ instead of pretending a background agent ran.
59
+
53
60
  ## Usage
54
61
 
55
62
  Open your AI coding tool in any project directory and type:
@@ -102,6 +109,16 @@ This reads `.godpowers/PROGRESS.md`, scans disk, reconciles any drift, and
102
109
  suggests the next logical command. The SessionStart hook does the same thing
103
110
  when you open a new session in a Godpowers project.
104
111
 
112
+ The same status engine is available from the installer CLI for humans, CI,
113
+ Codex, Claude, Cursor, Gemini, OpenCode, Windsurf, Antigravity, and any host
114
+ runtime that can execute Node:
115
+
116
+ ```bash
117
+ npx godpowers status --project=.
118
+ npx godpowers next --project=.
119
+ npx godpowers status --project=. --json
120
+ ```
121
+
105
122
  ### Slash Commands
106
123
 
107
124
  | Command | What it does | Spawns agent |
@@ -172,6 +189,14 @@ Under `--yolo`, Godpowers also auto-applies Pillars sync proposals when
172
189
  durable `.godpowers` artifacts change project truth. The decision is logged to
173
190
  `.godpowers/YOLO-DECISIONS.md`.
174
191
 
192
+ Every completing command now ends with a **Godpowers Dashboard**. It shows the
193
+ current phase, tier, step count, percent complete, PRD and roadmap visibility,
194
+ recent work, proactive checks, open items, and the single recommended next
195
+ action. `/god-status` and `/god-next` use the same shape so the project never
196
+ ends in a vague "done" state. The dashboard is backed by
197
+ `lib/dashboard.js`, so status and next-step commands share one executable
198
+ source of disk truth instead of parallel Markdown-only contracts.
199
+
175
200
  For existing codebases and org-constrained new projects, God Mode now runs a
176
201
  greenfield simulation audit and then actions it through a greenfieldification
177
202
  plan. It pauses before risky artifact rewrites because that process can change
package/RELEASE.md CHANGED
@@ -1,11 +1,10 @@
1
- # Godpowers 1.6.10 Release
1
+ # Godpowers 1.6.12 Release
2
2
 
3
3
  Date: 2026-05-16
4
4
 
5
- Godpowers 1.6.10 makes progress easier to understand while work is running and
6
- when it closes. The goal of this patch is to keep user-facing output from
7
- requiring internal "arc" vocabulary and to show PRD, roadmap, milestone, and
8
- completion position wherever status is reported.
5
+ Godpowers 1.6.12 turns progress visibility into executable product behavior.
6
+ The same dashboard logic now powers slash command closeouts, direct CLI status
7
+ checks, JSON status output, and next-action recommendations from disk state.
9
8
 
10
9
  ## What is stable
11
10
 
@@ -14,75 +13,70 @@ completion position wherever status is reported.
14
13
  - 13 executable workflows
15
14
  - 36 intent recipes
16
15
  - 15-runtime installer
17
- - Codex installs with 39 generated `god-*.toml` agent metadata files
18
- - Safe-sync routing before deploy, observe, harden, launch, or god-mode work
19
- - Critical harden finding gate before launch
16
+ - 453 package files in the npm tarball
17
+ - Codex installs with generated `god-*.toml` agent metadata files
18
+ - Markdown specialist agent contracts at `<runtime>/agents/god-*.md`
19
+ - Shared runtime bundle at `<runtime>/godpowers-runtime`
20
20
  - Native Pillars project context through `AGENTS.md` and `agents/*.md`
21
21
  - `.godpowers/` workflow state and artifact layout
22
- - Core schemas: intent, state, events, workflow, routing, recipes, extension
23
- manifests
24
- - Extension pack compatibility range for the 1.x line
25
- - Domain precision through `.godpowers/domain/GLOSSARY.md` and DG-01 through
26
- DG-05 checks
22
+ - Safe-sync routing before deploy, observe, harden, launch, or god-mode work
23
+ - Critical harden finding gate before launch
27
24
  - GSD-style proposition closeouts for exploratory, diagnostic, audit,
28
25
  lifecycle, status, reconciliation, and decision-support outputs
29
26
  - Plain-language project-run wording in user-facing reports
30
- - Planning visibility blocks for PRD, roadmap, milestone, and completion basis
31
27
 
32
28
  ## What is new
33
29
 
34
- - The core Godpowers skill now defines "arc" as internal vocabulary and tells
35
- user-facing output to prefer "project run", "workflow", "phase", "current
36
- step", or "current milestone".
37
- - `/god-mode`, `god-orchestrator`, `/god-status`, `/god-next`, and
38
- `/god-lifecycle` now include planning visibility guidance for PRD, roadmap,
39
- milestone, and percent complete when the information is available.
40
- - The installer, session-start hook, `/god` front door, routing descriptions,
41
- workflow descriptions, and high-traffic skill propositions now use
42
- project-run language instead of unexplained "arc" wording.
43
- - Checkpoint and session-start summaries display lifecycle `in-arc` as
44
- "in progress" while preserving the internal state key for compatibility.
45
-
46
- ## What 1.6.10 means
47
-
48
- Godpowers 1.6.10 does not expand the public command surface. It changes how
49
- Godpowers explains itself: the user should see the current project run, PRD,
50
- roadmap, milestone, completion percentage, open items, and next action without
51
- decoding internal terminology.
52
-
53
- Internal workflow names and state constants such as `full-arc.yaml` and
54
- `in-arc` remain unchanged for compatibility. Visible reports should translate
55
- them to plain language.
56
-
57
- Safe sync and unresolved Critical harden findings remain release-truth gates.
58
- Per-repo Quarterback ownership remains intact for Mode D suite work.
59
-
60
- ## Stability policy
61
-
62
- During the 1.x stability window, do not add broad new command families, change
63
- schema formats, or rename public artifacts without evidence from real use.
64
-
65
- The `v1.6.10` git tag points to the release commit that matches the npm
66
- `godpowers@1.6.10` package. Public publishes should prefer the tag-triggered
67
- GitHub workflow so npm provenance, git history, and release notes stay aligned.
68
-
69
- Allowed changes:
70
-
71
- - Critical bug fixes
72
- - Documentation clarity
73
- - Test coverage for frozen behavior
74
- - Compatibility fixes for supported AI coding tools
75
- - Small fixes that make existing 1.x behavior work as documented
76
-
77
- Deferred changes:
78
-
79
- - New lifecycle phases
80
- - New schema versions
81
- - Pillars format changes
82
- - Major routing semantics
83
- - Large extension API changes
84
-
85
- ## Adoption ask
86
-
87
- Run `npm run release:check` before publishing changes. Report any release path
88
- step that still depends on memory, manual inspection, or local machine state.
30
+ - Added `lib/dashboard.js`, a shared executable status engine.
31
+ - Added `godpowers status --project .`.
32
+ - Added `godpowers next --project .`.
33
+ - Added `--json` output for machine-readable status and next-step routing.
34
+ - Dashboard output now reports current phase, tier position, percentage,
35
+ planning visibility, proactive checks, open items, and recommended action.
36
+ - `/god-status`, `/god-next`, `/god-mode`, and `god-orchestrator` now reference
37
+ the same dashboard engine when local runtime execution is available.
38
+ - Tests now cover dashboard computation, rendering, CLI output, review queue
39
+ detection, and git porcelain parsing.
40
+
41
+ ## Platform behavior
42
+
43
+ Claude Code, Codex, Cursor, Windsurf, Gemini, OpenCode, Copilot, Augment,
44
+ Trae, Cline, Kilo, Antigravity, Qwen, CodeBuddy, and Pi all receive the same
45
+ portable Markdown agent contracts. Codex also receives `god-*.toml` files as
46
+ its registry adapter.
47
+
48
+ The dashboard engine ships in the installed runtime bundle so host tools can
49
+ use one shared implementation instead of parallel command-specific Markdown
50
+ contracts. If a host platform cannot provide a true fresh-context agent spawn,
51
+ Godpowers must say so visibly and report the work as local runtime only or
52
+ simulated in current context.
53
+
54
+ ## Safety policy
55
+
56
+ Godpowers may proactively suggest next steps and may run directly evidenced
57
+ local helpers. It may spawn bounded agents only when the current workflow owns
58
+ that surface.
59
+
60
+ Godpowers still must not auto-run these without explicit user intent:
61
+
62
+ - deployed staging verification against a guessed URL
63
+ - production launch
64
+ - provider dashboard, admin console, DNS, credential, or secret checks
65
+ - broad dependency upgrades
66
+ - destructive repair, rollback, reset, delete, or cleanup
67
+ - clearing `.godpowers/REVIEW-REQUIRED.md`
68
+ - accepting Critical security findings
69
+ - git stage, commit, push, package, release, or publish
70
+
71
+ ## Validation
72
+
73
+ Release validation includes:
74
+
75
+ - `node scripts/test-dashboard.js`
76
+ - `npm test`
77
+ - `npm run test:audit`
78
+ - `node scripts/check-package-contents.js`
79
+ - `git diff --check`
80
+
81
+ The `v1.6.12` git tag points to the release commit that matches the npm
82
+ `godpowers@1.6.12` package.
package/SKILL.md CHANGED
@@ -49,6 +49,19 @@ unresolved Critical security findings.
49
49
  Every execution agent gets a fresh context window. The orchestrator is thin; it
50
50
  spawns workers with full context budgets. This defeats context rot.
51
51
 
52
+ Spawning is platform-neutral. Use the host platform's native agent spawning
53
+ mechanism and the installed `agents/god-*.md` contract:
54
+ - Claude Code: spawn the matching Markdown agent from `~/.claude/agents/`.
55
+ - Codex: spawn the matching Codex agent type from `~/.codex/agents/*.toml`;
56
+ the Markdown copy remains the source contract.
57
+ - Cursor, Windsurf, Gemini, OpenCode, Copilot, Augment, Trae, Cline, Kilo,
58
+ Antigravity, Qwen, CodeBuddy, and Pi: use the platform's supported agent or
59
+ subagent mechanism against the installed Markdown files.
60
+
61
+ When a platform cannot spawn a true fresh-context agent, say so plainly,
62
+ preserve the same role contract, and report `Agent: none, local runtime only`
63
+ or `Agent: simulated in current context` in the visible auto-invoked card.
64
+
52
65
  ### 6. TDD Enforcement
53
66
  Tests are written before implementation. Code written before its test is flagged
54
67
  and must be rewritten. RED-GREEN-REFACTOR is not optional.
@@ -99,12 +112,28 @@ that edits code or artifacts, do not stop at "complete" plus validation. End
99
112
  with a disk-derived closeout that tells the user the current state and what is
100
113
  next.
101
114
 
115
+ Every closeout must include a **Godpowers Dashboard**. This dashboard is the
116
+ same mental model across `/god-status`, `/god-next`, `/god-mode`, and every
117
+ command that completes, pauses, or proposes work.
118
+
119
+ When the runtime bundle is available, compute this with
120
+ `lib/dashboard.compute(projectRoot)` and render it with
121
+ `lib/dashboard.render(result)`. The executable dashboard engine is the shared
122
+ source for phase, tier, step, progress, PRD and roadmap visibility, proactive
123
+ checks, open items, and the next route. If the runtime module is unavailable,
124
+ fall back to a manual disk scan and say `Dashboard engine: unavailable, manual
125
+ scan used`.
126
+
102
127
  Use this shape:
103
128
 
104
129
  ```
130
+ Godpowers Dashboard
131
+
105
132
  Current status:
106
133
  State: <complete | partial | blocked | complete with deferred item>
107
- Progress: <pct>% (<done> of <total> steps complete; current step <n> of <total>) when available
134
+ Phase: <plain-language phase> (tier <human ordinal> of <human total>) when available
135
+ Step: <sub-step label> (<step n> of <total steps>) when available
136
+ Progress: <pct>% (<done> of <total> steps complete) when available
108
137
  Worktree: <clean | modified files unstaged | staged changes | mixed>
109
138
  Index: <untouched | staged files listed>
110
139
 
@@ -121,8 +150,18 @@ What changed:
121
150
  Validation:
122
151
  + <command>: <result>
123
152
 
153
+ Proactive checks:
154
+ Checkpoint: <fresh | refreshed | missing | stale | conflicts with state.json>
155
+ Reviews: <none | N pending, suggest /god-review-changes>
156
+ Sync: <fresh | missing | stale | local helper ran | suggest /god-sync>
157
+ Docs: <fresh | possible drift, suggest /god-docs>
158
+ Runtime: <not-applicable | known URL, suggest /god-test-runtime | no known URL, defer deployed verification>
159
+ Security: <clear | sensitive files changed, suggest /god-harden>
160
+ Dependencies: <clear | dependency files changed, suggest /god-update-deps>
161
+ Hygiene: <fresh | stale, suggest /god-hygiene>
162
+
124
163
  Open items:
125
- 1. <deferred staging, unstaged files, pending review, or none>
164
+ 1. <deferred staging, unstaged files, pending review, blocker, or none>
126
165
 
127
166
  Next:
128
167
  Recommended: <one concrete command or user decision>
@@ -136,6 +175,11 @@ pre-existing unrelated changes, say the index was left untouched and recommend
136
175
  a scoped review or staging command rather than implying the project is fully
137
176
  shipped.
138
177
 
178
+ When the command only recommends work, keep the same dashboard but set
179
+ `State: proposal` and end with the proposition block from Section 9. When the
180
+ command pauses, set `State: blocked` or `State: paused` and make `Next` the
181
+ one exact user decision needed to continue.
182
+
139
183
  ### 11. User-Facing Vocabulary
140
184
  Godpowers may use internal words such as "arc" in routing, recipes, and agent
141
185
  implementation notes. Do not require the user to decode that word in visible
@@ -145,7 +189,7 @@ Use these plain-language substitutions in user-facing responses:
145
189
  - Say "project run" or "workflow" instead of "arc".
146
190
  - Say "phase" or "current step" instead of "tier" unless the user has asked
147
191
  for tier details. If tier detail helps, say both, for example
148
- "Planning, Tier 1".
192
+ "Planning phase, tier 2 of 4".
149
193
  - Say "current milestone" or "roadmap step" when ROADMAP.md has a matching
150
194
  milestone.
151
195
  - If you must use "arc", define it once as "the end-to-end project workflow"
@@ -156,6 +200,103 @@ and roadmap visibility when those files exist or are expected. Show whether
156
200
  the PRD is done, whether the roadmap exists, which milestone or tier is active,
157
201
  how close the tracked workflow is to completion, and the next concrete move.
158
202
 
203
+ ### 12. Auto-Invoke Visibility
204
+ When Godpowers automatically runs another command, agent, or local runtime
205
+ helper, show the user what happened. Do not describe these as "background"
206
+ unless they are truly detached from the current run. Most Godpowers sync work
207
+ is auto-invoked but still part of the current workflow.
208
+
209
+ Use this shape whenever an automatic step runs or is skipped:
210
+
211
+ ```
212
+ Auto-invoked:
213
+ Trigger: <what caused this automatic step>
214
+ Agent: <god-updater | god-context-writer | none, local runtime only>
215
+ Local syncs:
216
+ + <reverse-sync | pillars-sync | checkpoint-sync | context-refresh>: <result or skipped reason>
217
+ Artifacts: <changed files, no-op, or deferred>
218
+ Log: <SYNC-LOG.md, CHECKPOINT.md, REVIEW-REQUIRED.md, or none>
219
+ ```
220
+
221
+ If no agent was spawned, say so plainly:
222
+
223
+ ```
224
+ Agent: none, local runtime only
225
+ Why: this path calls lib/reverse-sync.run directly
226
+ ```
227
+
228
+ Automatic steps that especially need visible reporting:
229
+ - `/god-sync` spawning `god-updater`
230
+ - `god-updater` calling reverse-sync, Pillars sync, checkpoint sync, or
231
+ AI-tool context refresh
232
+ - `/god-mode` mandatory final sync
233
+ - standards checks between routed stages
234
+ - brownfield and bluefield automatic `/god-preflight`
235
+ - DESIGN/PRODUCT change detection that spawns `god-design-reviewer`
236
+ - `/god-scan` when it runs reverse-sync directly without an agent
237
+ - checkpoint refresh after state mutations
238
+
239
+ ### 13. Proactive Auto-Invoke Policy
240
+ Godpowers should be proactive from disk evidence, not from guesswork. Before
241
+ auto-invoking anything, classify the action by risk and apply the safest
242
+ allowed behavior.
243
+
244
+ #### Level 1: Auto-suggest, read-only
245
+ Run or apply these by default in every relevant closeout:
246
+ - Compute `/god-next` routing after successful commands.
247
+ - Show `/god-status` style progress after `/god-sync`, `/god-scan`, and
248
+ `/god-mode`.
249
+ - Suggest `/god-review-changes` when `.godpowers/REVIEW-REQUIRED.md` has
250
+ pending items.
251
+ - Suggest `/god-hygiene` after a full project run, after 30 days, or when
252
+ status shows stale docs, deps, or review queues.
253
+ - Suggest `/god-locate` when `.godpowers/CHECKPOINT.md` is missing, stale, or
254
+ conflicts with `state.json`.
255
+
256
+ #### Level 2: Auto-run local helpers, visible and logged
257
+ Run these local runtime helpers automatically when their trigger is present:
258
+ - `lib/checkpoint.syncFromState` after every `state.json` or
259
+ `PROGRESS.md` mutation.
260
+ - Lightweight reverse-sync or linkage scan after code or artifact edits.
261
+ - Pillars sync planning after durable artifact truth changes.
262
+ - Context refresh dry-run after `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`,
263
+ `.cursor/rules/`, `.windsurfrules`, `.github/copilot-instructions.md`,
264
+ `.clinerules`, `.roo/`, or `.continue/` changes.
265
+ - Progress recomputation after every command that changes artifacts.
266
+
267
+ #### Level 3: Auto-spawn scoped specialist agents
268
+ Spawn these agents only when the trigger is direct and scope is bounded:
269
+ - `god-design-reviewer` when `DESIGN.md` or `PRODUCT.md` changed.
270
+ - `god-updater` after feature, hotfix, refactor, build, deploy, observe,
271
+ launch, harden, docs, upgrade, or dependency workflows change code or
272
+ artifacts.
273
+ - `god-docs-writer` in drift-check mode when docs changed after code changed,
274
+ or code changed after docs that claim current behavior.
275
+ - `god-browser-tester` when frontend-visible files changed and a known local,
276
+ preview, staging, or production URL is evidenced.
277
+ - `god-harden-auditor` suggestion after security-sensitive files changed;
278
+ auto-spawn only inside `/god-harden`, `/god-hotfix`, `/god-launch`, or
279
+ `/god-mode`.
280
+ - `god-deps-auditor` suggestion after dependency files changed; auto-spawn only
281
+ inside `/god-update-deps`, `/god-hygiene`, or an explicitly approved project
282
+ workflow.
283
+
284
+ #### Level 4: Explicit approval required
285
+ Never auto-run these from inference alone:
286
+ - deployed staging verification against a guessed URL
287
+ - production launch
288
+ - provider dashboard, admin console, DNS, credential, or secret checks
289
+ - broad dependency upgrades
290
+ - destructive repair, rollback, reset, delete, or cleanup
291
+ - clearing `.godpowers/REVIEW-REQUIRED.md`
292
+ - accepting Critical security findings
293
+ - git stage, commit, push, package, release, or publish
294
+
295
+ Every auto-invoke decision must be explainable from one of these inputs:
296
+ changed files, Godpowers artifacts, `state.json`, `PROGRESS.md`,
297
+ `CHECKPOINT.md`, `SYNC-LOG.md`, `REVIEW-REQUIRED.md`, routing YAML, recipe YAML,
298
+ or explicit user intent.
299
+
159
300
  ---
160
301
 
161
302
  ## Operating Modes
@@ -134,7 +134,7 @@ For each operation, return to the spawning skill with:
134
134
 
135
135
  When you need work done IN a repo (version bump, patch slice, etc.),
136
136
  create a private handoff in that repo first, then spawn that repo's
137
- `god-orchestrator` via the Task tool with only a display-safe pointer.
137
+ `god-orchestrator` via the host platform's native agent spawning mechanism with only a display-safe pointer.
138
138
 
139
139
  Per-repo handoff path:
140
140
  `.godpowers/runs/<run-id>/COORDINATOR-ORCHESTRATOR-HANDOFF.md`