gaslighting-engine 0.2.1 → 0.3.0

package/examples/landing-page/.gaslighting/AGENT_RUNTIME.md

@@ -0,0 +1,54 @@
+# AGENT_RUNTIME.md
+
+This file defines how Codex must operate after Gaslighting-engine prepares the project.
+
+It is not a suggestion file.
+
+It is the runtime loop.
+
+## Mission
+
+- Original request: Build a landing page for a new service that captures leads.
+- Project type: landing_page
+- Classification confidence: high
+- Runtime stance: Codex-first, full-scope, no fake completion.
+
+## Required Loop
+
+1. Read the control files before implementation.
+2. Restate the project purpose in one short paragraph.
+3. Convert the PRD into a concrete task checklist.
+4. Implement the checklist systematically.
+5. Use scripts or structured data for repetitive work.
+6. Verify the result with build, tests, lint, browser checks, or direct inspection as appropriate.
+7. Update `.gaslighting/DECISION_LOG.md` when a product or technical decision changes.
+8. Update `.gaslighting/MISSING_INFO.md` when missing information is discovered or resolved.
+9. Update `.gaslighting/MEMORY.md` when stable project facts are learned.
+10. Update `.gaslighting/PROJECT_CARE.md` when GitHub, domain, deployment, env, analytics, email, or launch ownership changes.
+
+## Agent Rules
+
+- Do not ask the user for non-blocking information before starting work.
+- Do make a reasonable temporary assumption, document it, and continue.
+- Do not create only the visible surface and ignore operational details.
+- Do not stop at a plan when implementation was requested.
+- Do not call a scaffold complete.
+- Do not call TODO comments implementation.
+- Do not call examples full coverage.
+- Do not hide risk behind confident wording.
+- Do not over-engineer to look sophisticated.
+- Do not under-deliver because the work is repetitive.
+
+## Codex Execution Contract
+
+Before claiming completion, Codex must be able to answer:
+
+- What exact requested scope was completed?
+- What exact files changed?
+- What verification was run?
+- What assumptions are still active?
+- What missing information remains?
+- What care risks remain?
+- Is anything incomplete?
+
+If anything is incomplete, use the failure declaration protocol in `.gaslighting/GASLIGHTING.md`.

package/examples/landing-page/.gaslighting/CODEX_PROMPT.md

@@ -14,7 +14,8 @@ Read the following files before doing any work:
 6. `.gaslighting/DECISION_LOG.md`
 7. `AGENTS.md`
 8. `.gaslighting/MEMORY.md`
-9. `.gaslighting/PROJECT_CARE.md`
+9. `.gaslighting/AGENT_RUNTIME.md`
+10. `.gaslighting/PROJECT_CARE.md`
 
 Then implement the project MVP.
 
@@ -32,4 +33,10 @@ Rules:
 
 Before claiming completion, perform the self-audit in `.gaslighting/GASLIGHTING.md`.
 
+To refresh the Codex skill files later, run:
+
+```bash
+npx gaslighting-engine@latest upgrade
+```
+
 Proceed.

package/examples/landing-page/AGENTS.md

@@ -13,8 +13,9 @@ Before doing any work, read these files in order:
 5. `.gaslighting/ASSUMPTIONS.md`
 6. `.gaslighting/DECISION_LOG.md`
 7. `.gaslighting/MEMORY.md`
-8. `.gaslighting/PROJECT_CARE.md`
-9. `.gaslighting/AGENTS.md`
+8. `.gaslighting/AGENT_RUNTIME.md`
+9. `.gaslighting/PROJECT_CARE.md`
+10. `.gaslighting/AGENTS.md`
 
 Do not reduce scope.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gaslighting-engine",
-  "version": "0.2.1",
+  "version": "0.3.0",
   "description": "LUDGI Gaslighting-engine: a hardcore project-discipline generator for AI coding agents.",
   "type": "module",
   "license": "MIT",