GreyDGL · Jul 14, 2026 · Jul 14, 2026 · Jul 14, 2026 · Jul 14, 2026
diff --git a/AGENT.md b/AGENT.md
@@ -9,7 +9,8 @@ public project page and must not be edited unless the user explicitly requests t
 - `pentestgpt_agent/` — maintained autonomous Supervisor/Executor framework; nested uv project.
 - `pentestgpt_legacy/` — maintained human-driven USEN-2024-style client.
 - `../UnifedAgentWrapper/` — canonical `unified-agent` package used by the framework.
-- `../xbow-benchmark/` — benchmark harness and results; owned by a separate repository.
+- `../xbow-benchmark/` — reference-only benchmark harness and historical results; it is not a
+  supported `pentestgpt-agent` CLI, CI, or runtime integration.
 - `unified_agent/` — obsolete compatibility copy retained by root packaging/Docker only. Do not add
   features here or confuse it with the framework dependency.
 
@@ -54,9 +55,9 @@ make docker-login
 make docker-auth-status
 ```
 
-The root Docker image does not currently contain `pentestgpt_agent`; treat `make docker-run` and the
-external framework benchmark path as pending runtime wiring. Benchmark commands belong in
-`../xbow-benchmark`.
+The root Docker image does not currently contain `pentestgpt_agent`; treat `make docker-run` as
+pending runtime wiring. The sibling XBOW checkout is retained only as a reference and is not a
+product verification path.
 
 ## Documentation map
 

diff --git a/CLAUDE.md b/CLAUDE.md
@@ -9,7 +9,8 @@ public project page and must not be edited unless the user explicitly requests t
 - `pentestgpt_agent/` — maintained autonomous Supervisor/Executor framework; nested uv project.
 - `pentestgpt_legacy/` — maintained human-driven USEN-2024-style client.
 - `../UnifedAgentWrapper/` — canonical `unified-agent` package used by the framework.
-- `../xbow-benchmark/` — benchmark harness and results; owned by a separate repository.
+- `../xbow-benchmark/` — reference-only benchmark harness and historical results; it is not a
+  supported `pentestgpt-agent` CLI, CI, or runtime integration.
 - `unified_agent/` — obsolete compatibility copy retained by root packaging/Docker only. Do not add
   features here or confuse it with the framework dependency.
 
@@ -54,9 +55,9 @@ make docker-login
 make docker-auth-status
 ```
 
-The root Docker image does not currently contain `pentestgpt_agent`; treat `make docker-run` and the
-external framework benchmark path as pending runtime wiring. Benchmark commands belong in
-`../xbow-benchmark`.
+The root Docker image does not currently contain `pentestgpt_agent`; treat `make docker-run` as
+pending runtime wiring. The sibling XBOW checkout is retained only as a reference and is not a
+product verification path.
 
 ## Documentation map
 

diff --git a/Makefile b/Makefile
@@ -33,9 +33,6 @@ help:
 	@echo "  make docker-run        Pending: framework is not baked into the image yet"
 	@echo "  make docker-shell        Interactive shell in the tool container"
 	@echo "  make docker-nuke         Remove login volumes (forces re-login)"
-	@echo ""
-	@echo "Benchmarks moved to the sibling ../xbow-benchmark repo:"
-	@echo "  make -C ../xbow-benchmark help   (uses the pentestgpt:latest image built here)"
 
 # ============================================================================
 # Setup
@@ -193,7 +190,7 @@ docker-shell:
 # diagnostic instead of failing later with an opaque "command not found".
 docker-run:
 	@echo "pentestgpt_agent is not baked into $(DOCKER_IMAGE) yet."
-	@echo "Use 'make run TARGET=...' locally or the temporary isolated qualification runner."
+	@echo "Use 'make run TARGET=...' locally until a supported framework image is implemented."
 	@exit 2
 
 # Stop/remove the compose container but KEEP the login volumes (stay logged in)

diff --git a/README.md b/README.md
@@ -223,103 +223,16 @@ export LANGFUSE_ENABLED=false
 
 ---
 
-## Benchmarks
-
-PentestGPT achieved an **86.5% success rate** (90/104 benchmarks) on the XBOW validation suite
-(single-model, Dec-2025 baseline):
-
-- **Cost**: Average $1.11, Median $0.42 per successful benchmark
-- **Time**: Average 6.1 minutes, Median 3.3 minutes per successful benchmark
-- **Success rates by difficulty**:
-  - Level 1: 91.1%
-  - Level 2: 74.5%
-  - Level 3: 62.5%
-
-### Multi-model benchmarking
-
-The CTF pipeline is backend-pluggable (`--backend {claude,codex}`), enabling head-to-head comparison
-of Claude (`opus`/`sonnet`) and OpenAI Codex (`gpt-5.5`/`gpt-5.4-mini`) on the same challenges.
-
-The current comparable Docker baseline is
-[`benchmark/pilot/DOCKER_REPORT.md`](benchmark/pilot/DOCKER_REPORT.md): 10 representative medium/hard
-xbow targets, four model/runtime combinations, 40 runs total. The tool ran inside
-`pentestgpt:latest`, joined each target's Docker network, and attacked the target container IP
-directly. Raw records are in
-[`benchmark/pilot/docker_results.jsonl`](benchmark/pilot/docker_results.jsonl).
-The companion comparison report,
-[`benchmark/pilot/DOCKER_COMPARISON.md`](benchmark/pilot/DOCKER_COMPARISON.md), joins this pipeline
-baseline with the pure single-agent baseline and the current `pentestgpt_agent` framework smoke row.
-
-| Model | Solved | Solve rate | Avg solved time |
-|---|---:|---:|---:|
-| Codex `gpt-5.5` | 7/10 | 70% | 424s |
-| Claude Sonnet | 6/10 | 60% | 458s |
-| Claude Opus | 5/10 | 50% | 304s |
-| Codex `gpt-5.4-mini` | 3/10 | 30% | 427s |
-
-The earlier host-loopback pilot remains in [`benchmark/pilot/REPORT.md`](benchmark/pilot/REPORT.md)
-for historical comparison. The reusable Docker-network runner is
-[`benchmark/pilot/run_docker_bench.py`](benchmark/pilot/run_docker_bench.py); it supports
-`single`, `pipeline`, and `framework` variants.
-
-Quick Docker-network smoke:
+## Benchmark history
 
-```bash
-make docker-build
-make docker-login
-
-cd benchmark/pilot
-PILOT_COMBOS="claude-sonnet" uv run python run_docker_bench.py \
-  --variant pipeline \
-  --benchmarks XBEN-001-24,XBEN-068-24 \
-  --timeout 900 \
-  --concurrency 2 \
-  --out docker_parallel_smoke.jsonl
-```
-
-Run the next-gen `pentestgpt_agent` framework under the same harness:
-
-```bash
-make docker-bench-agent-quick
-```
-
-Run the full final best-of-three comparison:
+PentestGPT achieved an **86.5% success rate** (90/104 benchmarks) on an XBOW validation-suite
+experiment in December 2025. That number is a historical research result, not a current
+`pentestgpt-agent` regression guarantee.
 
-```bash
-make docker-bench-bestof3
-```
-
-That command runs a framework canary, then fresh `single`, `pipeline`, and `framework` sweeps over
-the 10-target suite with three attempts per cell. It writes timestamped JSONL files plus
-`benchmark/pilot/DOCKER_BEST_OF_3_<run-id>.md`. If infra/auth/liveness rows leave cells incomplete,
-rerun with the same `BOF3_RUN_ID` to fill retryable attempts.
-
-The Docker runner is host-driven: it builds each XBOW fixture if needed, starts the target with
-Docker Compose, discovers the target container's internal Docker-network URL, runs one fresh
-`pentestgpt:latest` tool container on that network, writes one JSONL row, and tears the target down.
-It now defaults to each fixture's normal Compose project name because XBOW images are pre-built under
-those names. `--concurrency N` parallelizes across distinct benchmarks while model combos remain
-serial inside each benchmark; `docker compose up` uses `--no-build` and Docker preflight fails fast if
-the daemon or tool image is unavailable. Fixed host `ports:` are automatically serialized when they
-overlap. Codex cells default to `--max-codex 1` because shared ChatGPT OAuth refresh tokens can race;
-only use `--max-codex 0` with auth that is safe to refresh concurrently.
-
-Remaining benchmark work is the full 59-target medium/hard sweep and cost normalization. Start real
-sweeps at `--concurrency 2`, move to `4` only after a small matrix is clean, and use
-`make docker-bench-bestof3` for the final 10-target best-of-three comparison. See
-[`benchmark/pilot/README.md`](benchmark/pilot/README.md) and
-[`docs/docker-dev-plan.md`](docs/docker-dev-plan.md).
-
-`make docker-bench-agent-quick` is the default local verification after changing `pentestgpt_agent/`.
-It runs Claude Sonnet on the known framework-smoke target `XBEN-080-24`, mounts the current checkout
-into the tool container with `--source-root`, uses framework executor concurrency `2`, appends a
-timestamped attempt so it does not skip stale rows, and writes
-`benchmark/pilot/docker_framework_quick.jsonl`. Override with Make variables for broader checks, for
-example:
-
-```bash
-make docker-bench-agent-quick AGENT_BENCH_TARGETS=XBEN-001-24,XBEN-068-24 AGENT_BENCH_CONCURRENCY=2 AGENT_BENCH_TIMEOUT=2400
-```
+XBOW harnesses and result archives are maintained outside this product repository as reference-only
+research artifacts. The supported PentestGPT CLI, Makefile, CI, and Docker runtime do not expose an
+XBOW runner. A future evaluation may reuse that corpus through a separately owned adapter without
+making it a product dependency.
 
 ---
 

diff --git a/docs/architecture.md b/docs/architecture.md
@@ -11,11 +11,12 @@ public project page and is intentionally being revised separately.
 PentestGPT_Project/
 ├── PentestGPT/          # framework, legacy interactive client, and tool image
 ├── UnifedAgentWrapper/  # canonical unified-agent package
-└── xbow-benchmark/      # Docker-network benchmark harness and results
+└── xbow-benchmark/      # reference-only benchmark harness and historical results
 ```
 
-The three directories are independent Git repositories. Benchmark code and result archives belong
-in `xbow-benchmark`; new benchmark logic must not be added here.
+The three directories are independent Git repositories. The XBOW checkout is retained as a
+reference corpus only; the product CLI, runtime, and CI do not depend on it. Benchmark logic and
+result archives must not be added here.
 
 ## Maintained runtime
 
@@ -107,10 +108,8 @@ deterministic strategy projection and duplicate/branch policy before adding anot
 - Local framework development runs from `pentestgpt_agent/`.
 - The root Docker image supplies pentest tools, provider CLIs, legacy code, and persisted auth. It
   does not currently bake in the maintained framework.
-- The temporary `pentestgpt_agent/scripts/run_xbow.py` qualification runner builds framework and
-  wrapper wheels before installing them into an isolated tool container. New benchmark development
-  belongs in the sibling `xbow-benchmark` repo; transfer this runner there when its framework path is
-  rewired.
+- The sibling `xbow-benchmark` checkout is a historical/reference artifact, not a supported runtime
+  or verification path. PentestGPT owns no XBOW runner.
 - HTB execution happens only on the authorized remote attack box described in the parent project
   guide, never directly from the development Mac.
 
@@ -119,6 +118,5 @@ deterministic strategy projection and duplicate/branch policy before adding anot
 1. Preserve the two-role loop and deterministic memory authority.
 2. Improve Supervisor coverage retrieval and convergence using saved-trace replay tests.
 3. Keep both roles fully enabled inside a restricted deployment environment.
-4. Move the remaining XBOW qualification runner to the benchmark repository.
-5. Remove the root `unified_agent` compatibility copy in a coordinated public-package cleanup.
-6. Add a goal-specific verifier only when non-CTF completion semantics require it.
+4. Remove the root `unified_agent` compatibility copy in a coordinated public-package cleanup.
+5. Add a goal-specific verifier only when non-CTF completion semantics require it.
diff --git a/docs/docker-dev-plan.md b/docs/docker-dev-plan.md
@@ -20,20 +20,19 @@ It deliberately excludes benchmark fixtures/results, credentials, workspaces, an
 
 The maintained `pentestgpt_agent` nested project is **not baked into this image**. Consequently,
 `make docker-run` deliberately fails fast with a wiring diagnostic instead of invoking an absent
-CLI. The temporary qualification runner in `pentestgpt_agent/scripts/run_xbow.py` works around this
-by building the framework and external wrapper wheels on the host, then installing those wheels into
-an isolated container environment.
+CLI. There is no product-owned benchmark workaround; framework-image wiring remains an independent
+deployment task.
 
 ## Repository ownership
 
 ```text
 PentestGPT/          image, auth helpers, framework source, legacy client
 UnifedAgentWrapper/  canonical provider-wrapper package
-xbow-benchmark/      Docker-network targets, scheduling, scoring, reports
+xbow-benchmark/      reference-only benchmark harness and historical results
 ```
 
-The benchmark harness no longer belongs under this repository. Run benchmark commands from
-`../xbow-benchmark`; do not add result JSONL or target orchestration here.
+The product does not support or invoke the sibling benchmark harness. Keep result JSONL, target
+orchestration, and benchmark-specific adapters out of this repository.
 
 ## Persistent provider login
 
@@ -89,7 +88,7 @@ There are two reasonable future shapes:
 2. Publish both packages and install pinned releases during the Docker build.
 
 Do not copy the root `unified_agent/` directory into the maintained framework. It is version 0.1-era
-compatibility code; the agent requires the pinned external 0.2 package.
+compatibility code; the agent requires the pinned external 0.3 package.
 
 Whichever shape is selected must make these checks true in a fresh container:
 
@@ -112,13 +111,9 @@ These are independent of framework design:
 - remove the root `unified_agent` copy and its SDK dependencies when the public-package cleanup is
   performed.
 
-## Benchmark handoff
+## External benchmark reference
 
-The sibling benchmark repository is host-driven: it starts each authorized XBOW fixture, attaches a
-fresh tool container to the target network, records one result, and tears the target down. Claude
-jobs may share their token volume; Codex jobs should remain serialized unless refresh-token safety is
-demonstrated.
-
-Its `framework` variant is still pending rewire to the renamed `pentestgpt_agent`. The working wheel
-build/install sequence in `pentestgpt_agent/scripts/run_xbow.py` is the implementation reference for
-that transfer.
+The sibling `xbow-benchmark` repository is retained only for historical reference. The product
+Makefile, CLI, CI, and Docker runtime do not invoke it. If a future evaluation reuses that corpus,
+design the adapter in an external evaluation repository rather than restoring a benchmark runner
+inside PentestGPT.
diff --git a/pentestgpt_agent/.gitignore b/pentestgpt_agent/.gitignore
@@ -8,6 +8,3 @@ __pycache__/
 build/
 dist/
 runs/*
-!runs/xben080-opus48-20260711/
-runs/xben080-opus48-20260711/*
-!runs/xben080-opus48-20260711/REPORT.md
diff --git a/pentestgpt_agent/CONTEXT.md b/pentestgpt_agent/CONTEXT.md
@@ -89,6 +89,6 @@ replace selection, but it must return canonical IDs for deterministic resolution
 ## Qualification semantics
 
 `RunStatus.COMPLETED` establishes structural completion and explicit evidence citation. It does not
-prove arbitrary semantic goal entailment. XBOW's held-out oracle supplies that final predicate; a
-general deployment needs a goal-specific verifier. Such a verifier should remain separate from
-scheduling and memory ownership.
+prove arbitrary semantic goal entailment. A held-out benchmark oracle can supply that final
+predicate; a general deployment needs a goal-specific verifier. Such a verifier should remain
+separate from scheduling and memory ownership.
diff --git a/pentestgpt_agent/README.md b/pentestgpt_agent/README.md
@@ -133,7 +133,7 @@ writes are reported as diagnostics rather than failures.
 
 ## UnifiedAgent dependency
 
-`unified-agent==0.2.0` is pinned to a public Git commit in `pyproject.toml`. It is the provider seam
+`unified-agent==0.3.0` is pinned to the merged inspector commit in `pyproject.toml`. It is the provider seam
 for Claude Code and Codex: task invocation, model/effort configuration, structured output, and
 normalized events. PentestGPT keeps task policy, memory, evidence, and completion outside that
 dependency.
@@ -148,8 +148,8 @@ episode a stable `<pentest-run-id>--<episode-id>` identity and attaches the role
 task, and attempt as metadata. The framework's trace journal remains canonical; the inspector
 database is an additional real-time diagnostic projection.
 
-During joint wrapper/application development, opt into the sibling 0.3 checkout without changing
-the release pin:
+To inspect local wrapper changes before publishing a new immutable dependency revision, explicitly
+opt into the sibling checkout:
 
 ```bash
 export PYTHONPATH="$(cd ../../UnifedAgentWrapper && pwd)"
@@ -167,24 +167,17 @@ PYTHONPATH=. .venv/bin/python -m unified_agent.monitor_web \
   --open
 ```
 
-The temporary XBOW qualification runner builds both local wheels and always writes its inspector
-database to `<artifact-dir>/agent-data/monitor.sqlite3`; it prints that path before target setup.
-Before merge, replace the current 0.2 dependency source with the reviewed immutable UnifiedAgent
-0.3 commit and regenerate `uv.lock`.
+The pinned 0.3 dependency provides the same monitor behavior in normal synchronized environments;
+the sibling override is only for wrapper development.
 
-## Benchmarks and qualifications
+## External evaluation references
 
-The Docker-network benchmark harness and result corpus live in the sibling `../../xbow-benchmark`
-repository. Do not add new benchmark orchestration here. `scripts/run_xbow.py` is a temporary
-qualification runner retained until its working framework-wheel path is transferred to that repo.
+The sibling `../../xbow-benchmark` checkout is retained only as a historical/reference corpus.
+There is no supported XBOW command, runner, or CI path in `pentestgpt-agent`, and benchmark-specific
+orchestration must stay outside this product repository. Historical engineering records may still
+name earlier XBEN qualifications; they are not current regression guarantees.
 
-Historical accepted local qualifications at PentestGPT commit `9e6d2c4`:
-
-- XBEN-068 command injection: solved in 4 tasks / 9 episodes, $0.856196, 238.9 seconds.
-- XBEN-080 Jinja2 SSTI: solved in 4 tasks / 9 episodes, $0.834787, 180.0 seconds.
-
-Those runs used the earlier audit contract; their numbers are historical evidence, not current
-regression guarantees. The later HTB Enigma qualification did not solve the target; see
+The later HTB Enigma qualification did not solve the target; see
 [`HTB_ENIGMA_QUALIFICATION_20260712.md`](HTB_ENIGMA_QUALIFICATION_20260712.md).
 
 ## Current limits

diff --git a/pentestgpt_agent/pyproject.toml b/pentestgpt_agent/pyproject.toml
@@ -8,17 +8,17 @@ version = "0.1.0"
 description = "A small, memory-backed penetration-testing agent built on UnifiedAgent."
 requires-python = ">=3.12,<4.0"
 dependencies = [
-    "unified-agent==0.2.0",
+    "unified-agent==0.3.0",
 ]
 
 [project.scripts]
 pentestgpt-agent = "pentestgpt_agent.trial:main"
 pentestgpt-agent-audit = "pentestgpt_agent.audit:main"
 
 [project.optional-dependencies]
-claude = ["unified-agent[claude]==0.2.0"]
-codex = ["unified-agent[codex]==0.2.0"]
-all = ["unified-agent[all]==0.2.0"]
+claude = ["unified-agent[claude]==0.3.0"]
+codex = ["unified-agent[codex]==0.3.0"]
+all = ["unified-agent[all]==0.3.0"]
 
 [dependency-groups]
 dev = [
@@ -29,7 +29,7 @@ dev = [
 ]
 
 [tool.uv.sources]
-unified-agent = { git = "https://github.com/PentestGPT-Project/UnifedAgentWrapper.git", rev = "d05d21f500f42c70847d8426dee1a2d5da9b59fa" }
+unified-agent = { git = "https://github.com/PentestGPT-Project/UnifedAgentWrapper.git", rev = "acff8eeadf93e367d4a1578d3a2739cbc9d3ace5" }
 
 [tool.uv]
 prerelease = "allow"