Build CompText Universe inside Hugging Face Space

hf_space/README.md

@@ -1,38 +1,54 @@
 ---
-title: CompText Prompt Compression Lab
-emoji: 🗜️
-colorFrom: blue
-colorTo: purple
+title: CompText Universe
+emoji: 🧭
+colorFrom: indigo
+colorTo: blue
 sdk: gradio
+python_version: 3.10.13
 sdk_version: 5.44.1
 app_file: app.py
+fullWidth: true
+header: mini
 pinned: false
 license: apache-2.0
+short_description: Explore CompText architecture, contracts and safe context compression.
+models:
+  - microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank
+tags:
+  - context-engineering
+  - prompt-compression
+  - software-engineering
+  - gradio
+preload_from_hub:
+  - microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank
 ---
 
-# CompText Prompt Compression Lab
+# CompText Universe
 
-A CPU-friendly Hugging Face Space for evaluating prompt and context compression with Microsoft LLMLingua-2.
+> Models are providers. Context is the product. Evidence is the trust layer. CompText is the kernel.
 
-## Features
+A public, experimental showcase for the CompText local engineering-orchestration architecture. The real CompText runtime remains local; this Space visualizes architecture and contracts, builds non-executable AIR and simulated Evidence previews, and tests fail-closed context compression.
 
-- Compress arbitrary prompts at configurable retention rates
-- Compare original and compressed text
-- Measure token reduction and runtime
-- Check preservation of negations, CLI flags, file paths, JSON keys, version numbers, and code-like symbols
-- Run a built-in benchmark suite
-- Export results as JSON
+## Included surfaces
 
-## Default model
+- Static seven-layer architecture explorer
+- Explicit capability maturity matrix
+- Workspace skill and safety-boundary explorer
+- Fail-closed hybrid LLMLingua-2 compression
+- Deterministic, non-executable AIR previews
+- Simulated, non-persistent Evidence previews
+- CompText-specific benchmark and JSON exports
 
-`microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank`
+## Hard boundaries
 
-The first startup can take several minutes because the model must be downloaded.
+- No provider calls
+- No repository writes
+- No runtime GitHub access
+- No API keys or runtime secrets
+- No AIR execution
+- No persistent prompt storage
+- No claim that planned or scaffolded components are production-ready
 
-## Hardware
+## Runtime
 
-Designed for Hugging Face Spaces `CPU Basic` (2 vCPU, 16 GB RAM, 50 GB ephemeral storage).
-
-## Safety
-
-This Space does not call external LLM APIs, require runtime secrets, access private repositories, or modify repositories.
+Only compression callbacks request ZeroGPU. Universe navigation, preview construction, contract views and secret-pattern checks are CPU-only and deterministic. The model is preloaded from the Hub during build to reduce first-request latency.

hf_space/app.py

@@ -10,51 +10,77 @@
 
 from compression import DEFAULT_MODEL, compress_text
 from metrics import summarize_checks
+from previews import build_air_preview, build_evidence_preview, scan_secrets
 from safety_checks import run_safety_checks
+from universe import capabilities_frame, contracts_frame, layers_frame, load_universe, overview_markdown, skills_frame
 
 ROOT = Path(__file__).parent
 BENCHMARK_CASES = json.loads((ROOT / "benchmark_cases.json").read_text(encoding="utf-8"))
+UNIVERSE = load_universe()
 EXAMPLE_TEXT = (
-    "Ändere AGENTS.md nicht und arbeite nur mit --dry-run. "
-    "CompText contains a detailed orchestration layer that coordinates providers, context preparation, policy checks, and output validation. "
-    "The architecture explanation is intentionally verbose so that the safe hybrid compressor has natural-language prose to reduce while preserving every technical constraint unchanged."
+    "Führe keine Live-Provider-Aufrufe aus. Ändere AGENTS.md nicht. "
+    "Analysiere ausführlich, wie CompText Context Packs für lokale Softwareentwicklung vorbereitet. "
+    "Nutze modules/cli/cli_entrypoint.py nur lesend mit --dry-run und gib einen JSON-Bericht zurück."
 )
 
 
-def _export(payload: object) -> str:
+def _json_export(payload: object) -> str:
     with NamedTemporaryFile("w", encoding="utf-8", suffix=".json", delete=False) as handle:
         json.dump(payload, handle, ensure_ascii=False, indent=2)
         return handle.name
 
 
 @spaces.GPU
 def compress_ui(text: str, retention_percent: int):
+    secret_matches = scan_secrets(text)
+    if secret_matches:
+        payload = {
+            "decision": "blocked",
+            "reason": "Potential secret material detected. Input was not processed.",
+            "secret_matches": secret_matches,
+        }
+        return text, payload, pd.DataFrame(), json.dumps(payload, ensure_ascii=False, indent=2), None
+
     result = compress_text(text, retention_rate=retention_percent / 100)
     checks = run_safety_checks(result.original_text, result.candidate_text)
     summary = summarize_checks(checks)
+    decision = "accepted" if result.accepted else "fallback"
     metrics = {
-        "Decision": "ACCEPTED" if result.accepted else "FALLBACK TO ORIGINAL",
-        "Fallback reason": result.fallback_reason or "",
+        "Decision": decision,
         "Original tokens": result.origin_tokens,
         "Output tokens": result.compressed_tokens,
-        "Net token reduction": f"{result.token_reduction_percent}%",
+        "Net reduction": f"{result.token_reduction_percent}%",
         "Protected segments": result.protected_segments,
         "Compressed segments": result.compressed_segments,
+        "Candidate safety": f"{summary['score_percent']}%",
         "Runtime": f"{result.runtime_seconds}s",
-        "Candidate safety": f"{summary['passed']}/{summary['total']}",
+        "Reason": result.fallback_reason or "",
     }
     rows = [
         {
-            "Check": c.name,
-            "Relevant": "Yes" if c.relevant else "No",
-            "Passed": "Yes" if c.passed else "No",
-            "Expected": ", ".join(c.expected),
-            "Missing": ", ".join(c.missing),
+            "Check": check.name,
+            "Relevant": "Yes" if check.relevant else "No",
+            "Passed": "Yes" if check.passed else "No",
+            "Expected": ", ".join(check.expected),
+            "Missing": ", ".join(check.missing),
         }
-        for c in checks
+        for check in checks
     ]
-    payload = {"compression": result.to_dict(), "candidate_safety": summary}
-    return result.compressed_text, metrics, pd.DataFrame(rows), json.dumps(payload, ensure_ascii=False, indent=2), _export(payload)
+    payload = {
+        "decision": decision,
+        "compression": result.to_dict(),
+        "safety": summary,
+        "air_preview": build_air_preview(text, {"decision": decision, "metrics": metrics}),
+        "evidence_preview": build_evidence_preview(text, result.compressed_text, decision, metrics),
+    }
+    return result.compressed_text, metrics, pd.DataFrame(rows), json.dumps(payload, ensure_ascii=False, indent=2), _json_export(payload)
+
+
+def preview_contracts(text: str):
+    matches = scan_secrets(text)
+    air = build_air_preview(text)
+    evidence = build_evidence_preview(text, text, "preview_only", {"secret_matches": matches})
+    return air, evidence
 
 
 @spaces.GPU
@@ -80,39 +106,92 @@ def run_benchmark(retention_percent: int):
             })
         except Exception as exc:
             rows.append({"Case": case["name"], "Category": case["category"], "Decision": "error", "Reason": str(exc)})
-    return pd.DataFrame(rows), _export(rows)
+    return pd.DataFrame(rows), _json_export(rows)
 
 
-with gr.Blocks(title="CompText Prompt Compression Lab") as demo:
+with gr.Blocks(title="CompText Universe") as demo:
+    gr.Markdown(overview_markdown(UNIVERSE))
     gr.Markdown(
-        "# 🗜️ CompText Safe Hybrid Compression Lab\n"
-        "Critical instructions, flags, paths, JSON, versions, and code symbols are protected. "
-        "Only natural-language segments are compressed. Unsafe or ineffective candidates automatically fall back to the original text."
+        "**Experimental public showcase:** no provider calls, no repository writes, no AIR execution, "
+        "no persistent prompt storage. The local CompText kernel remains the product runtime."
     )
-    with gr.Tab("Single prompt"):
-        input_text = gr.Textbox(label="Original prompt", value=EXAMPLE_TEXT, lines=14)
+
+    with gr.Tab("Overview"):
+        with gr.Row():
+            gr.Markdown(
+                "## Context is the product\n"
+                "CompText treats models as interchangeable providers while context, contracts and Evidence "
+                "form the durable engineering system. This Space visualizes that model without executing it."
+            )
+            gr.JSON(value={
+                "product_status": UNIVERSE["product"]["status"],
+                "snapshot_version": UNIVERSE["snapshot_version"],
+                "runtime_github_access": UNIVERSE["source"]["runtime_github_access"],
+                "compression_model": DEFAULT_MODEL,
+            }, label="Snapshot identity")
+        gr.Dataframe(value=contracts_frame(UNIVERSE), label="Core contracts", interactive=False)
+
+    with gr.Tab("Architecture"):
+        gr.Markdown(
+            "## Seven-layer model\n"
+            "`User → Terminal OS / UI → Runtime / Gateway / Agent Bus → AIR / Evidence / Memory → Provider Router`"
+        )
+        gr.Dataframe(value=layers_frame(UNIVERSE), label="Architecture layers", interactive=False)
+
+    with gr.Tab("Capabilities"):
+        gr.Markdown("## What exists, what is experimental, and what remains future")
+        gr.Dataframe(value=capabilities_frame(UNIVERSE), label="Capability matrix", interactive=False)
+
+    with gr.Tab("Compression Lab"):
+        gr.Markdown(
+            "## Fail-closed hybrid compression\n"
+            "Critical instructions, flags, paths, structured data and code-like symbols are protected. "
+            "Unsafe or unhelpful candidates return the exact original input."
+        )
+        input_text = gr.Textbox(label="Engineering context", value=EXAMPLE_TEXT, lines=14)
         retention = gr.Slider(10, 100, value=60, step=5, label="Retention rate (%)")
-        button = gr.Button("Compress safely", variant="primary")
-        compressed = gr.Textbox(label="Safe output", lines=14)
-        metrics = gr.JSON(label="Decision and metrics")
-        checks = gr.Dataframe(label="Candidate safety checks", interactive=False)
-        raw = gr.Code(label="Raw result", language="json")
-        download = gr.File(label="Download JSON result")
-        button.click(compress_ui, [input_text, retention], [compressed, metrics, checks, raw, download])
-    with gr.Tab("Benchmark suite"):
+        compress_button = gr.Button("Analyze and compress", variant="primary")
+        compressed = gr.Textbox(label="Final output", lines=14)
+        compression_metrics = gr.JSON(label="Decision and metrics")
+        checks = gr.Dataframe(label="Relevant preservation checks", interactive=False)
+        raw = gr.Code(label="Compression + AIR + Evidence payload", language="json")
+        download = gr.File(label="Download JSON")
+        compress_button.click(compress_ui, [input_text, retention], [compressed, compression_metrics, checks, raw, download])
+
+    with gr.Tab("AIR & Evidence"):
+        gr.Markdown(
+            "## Non-executable contract previews\n"
+            "AIR describes intended work. Evidence describes observed work. Here, AIR is always disabled and "
+            "Evidence is always marked as simulated."
+        )
+        preview_text = gr.Textbox(label="Engineering task", value=EXAMPLE_TEXT, lines=10)
+        preview_button = gr.Button("Build previews")
+        air_output = gr.JSON(label="AIR preview")
+        evidence_output = gr.JSON(label="Simulated Evidence preview")
+        preview_button.click(preview_contracts, preview_text, [air_output, evidence_output])
+
+    with gr.Tab("Skills"):
+        gr.Markdown("## Skill-grounded local workflow and safety boundaries")
+        gr.Dataframe(value=skills_frame(UNIVERSE), label="Workspace skills", interactive=False)
+
+    with gr.Tab("Benchmarks"):
+        gr.Markdown("## CompText-specific protected-context benchmark")
         bench_rate = gr.Slider(10, 100, value=60, step=5, label="Retention rate (%)")
-        bench_button = gr.Button("Run safe hybrid benchmark", variant="primary")
+        bench_button = gr.Button("Run benchmark", variant="primary")
         bench_table = gr.Dataframe(label="Benchmark results", interactive=False)
         bench_download = gr.File(label="Download benchmark JSON")
         bench_button.click(run_benchmark, bench_rate, [bench_table, bench_download])
-    with gr.Accordion("Decision policy", open=False):
+
+    with gr.Accordion("Model, provenance and limitations", open=False):
+        gr.JSON(value=UNIVERSE["source"], label="Static snapshot provenance")
         gr.Markdown(
-            f"**Model:** `{DEFAULT_MODEL}`\n\n"
-            "A result is accepted only when all relevant protected elements survive and net token reduction is at least 10%. "
-            "Otherwise the exact original input is returned."
+            f"**Compression model:** `{DEFAULT_MODEL}`\n\n"
+            "The Universe snapshot is committed data, not a live GitHub view. Safety checks are deterministic "
+            "preservation heuristics, not a semantic equivalence proof."
         )
 
+
 demo.queue(default_concurrency_limit=1, max_size=8)
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(ssr_mode=False)

hf_space/data/universe_snapshot.json

@@ -0,0 +1,65 @@
+{
+  "snapshot_version": "1.0",
+  "product": {
+    "name": "CompText",
+    "repository_name": "comptext",
+    "status": "local-dry-run-mvp",
+    "claim": "Models are providers. Context is the product. Evidence is the trust layer. CompText is the kernel.",
+    "description": "A local AI orchestration platform for software engineering. This public Space is an experimental, non-executing showcase."
+  },
+  "source": {
+    "repository": "ProfRandom92/Comptext",
+    "mode": "committed-static-snapshot",
+    "runtime_github_access": false,
+    "allowed_sources": [
+      "AGENTS.md",
+      "docs/COMPTEXT_ARCHITECTURE_v1.md",
+      ".agents/skills/comptext-local-autonomy/SKILL.md",
+      ".agents/skills/comptext-local-verify/SKILL.md",
+      ".agents/skills/comptext-workspace-validation/SKILL.md",
+      ".agents/skills/comptext-status/SKILL.md",
+      ".agents/skills/workspace-state/SKILL.md"
+    ]
+  },
+  "boundaries": [
+    "No provider calls",
+    "No repository writes",
+    "No runtime GitHub access",
+    "No secrets or environment access",
+    "No AIR execution",
+    "No persistent user input storage"
+  ],
+  "layers": [
+    {"id":"ui","name":"Terminal OS / UI","status":"scaffolded","purpose":"Human-facing workbench for sessions, workspaces, commands, providers, evidence and run queues.","inputs":"user intent","outputs":"normalized commands and views"},
+    {"id":"runtime","name":"Runtime","status":"scaffolded","purpose":"Coordinates runs, plans, verification, retries, queues and replay.","inputs":"AIR plans","outputs":"run state and events"},
+    {"id":"gateway","name":"Gateway","status":"planned","purpose":"Normalizes local provider-compatible message and response routes.","inputs":"local API requests","outputs":"normalized traffic"},
+    {"id":"agent-bus","name":"Agent Bus","status":"planned","purpose":"Coordinates specialized agents as explicit tasks with approval gates.","inputs":"tasks and roles","outputs":"task results and evidence"},
+    {"id":"air","name":"AIR","status":"scaffolded","purpose":"Describes intended work before execution: goal, context, tools, constraints, permissions and outputs.","inputs":"intent and context","outputs":"non-executed plan contract"},
+    {"id":"evidence","name":"Evidence","status":"scaffolded","purpose":"Records what actually happened without secrets, raw provider payloads or hidden reasoning.","inputs":"runtime and validation events","outputs":"redacted evidence events"},
+    {"id":"memory","name":"Memory / Knowledge Graph","status":"future","purpose":"Structured workspace knowledge spanning files, functions, tests, runs and evidence.","inputs":"validated workspace state","outputs":"retrievable context relationships"}
+  ],
+  "capabilities": [
+    {"name":"Local status","surface":"CLI","status":"implemented","network":"no","provider":"no","mutating":"no"},
+    {"name":"Local verification","surface":"CLI","status":"implemented","network":"no","provider":"no","mutating":"no"},
+    {"name":"Workspace schema validation","surface":"CLI","status":"implemented","network":"no","provider":"no","mutating":"no"},
+    {"name":"Hybrid context compression","surface":"HF Space","status":"experimental","network":"model-cache","provider":"no","mutating":"no"},
+    {"name":"AIR preview","surface":"HF Space","status":"experimental","network":"no","provider":"no","mutating":"no"},
+    {"name":"Simulated Evidence preview","surface":"HF Space","status":"experimental","network":"no","provider":"no","mutating":"no"},
+    {"name":"Provider Router","surface":"Kernel","status":"scaffolded","network":"disabled","provider":"not_configured","mutating":"no"},
+    {"name":"Workspace reflection runtime","surface":"Kernel","status":"future","network":"no","provider":"no","mutating":"local"},
+    {"name":"Autonomous PR merge","surface":"Agent workflow","status":"disabled","network":"yes","provider":"no","mutating":"yes"}
+  ],
+  "skills": [
+    {"name":"comptext-local-autonomy","purpose":"One-unit-at-a-time offline development loop.","validation":"python -m pytest; git diff --check","boundary":"No network, providers, secrets, GitHub writes or servers."},
+    {"name":"comptext-local-verify","purpose":"Verify status, subagents, workspace validation and doctor diagnostics.","validation":"comptext verify --dry-run","boundary":"Offline dry-run checks only."},
+    {"name":"comptext-workspace-validation","purpose":"Validate committed workspace examples against strict schemas.","validation":"comptext validate workspace --dry-run","boundary":"No active generation, network, databases or providers."},
+    {"name":"comptext-status","purpose":"Show file presence and local diagnostic state.","validation":"comptext status --dry-run","boundary":"Offline local-only data collection."},
+    {"name":"workspace-state","purpose":"Define future snapshot, delta and reflection-gate concepts.","validation":"schema and fixture validation","boundary":"No active runtime or interpretability claims."}
+  ],
+  "contracts": [
+    {"name":"AIR Plan","kind":"intent","fields":"version, intent, goal, context, files, tools, permissions, expected_outputs, metadata","execution":"never in this Space"},
+    {"name":"Evidence Event","kind":"observed-event","fields":"event_id, run_id, type, actor, tool, summary, hashes, timestamp, redaction","execution":"simulated preview only"},
+    {"name":"Run Record","kind":"run-index","fields":"run_id, air_hash, status, start_time, end_time, event_hashes, metrics","execution":"not created in this Space"},
+    {"name":"Workspace Snapshot","kind":"workspace-state","fields":"schema-defined local state","execution":"static concept preview only"}
+  ]
+}