diff --git a/CHANGELOG.md b/CHANGELOG.md
index b2d76b7..1ea2fab 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -12,6 +12,11 @@ Contributors add user-facing entries under `[Unreleased]` in the same PR. Mainta
 
 - **Citation:** Zenodo concept DOI `10.5281/zenodo.21552745` in `CITATION.cff`, README badge/Citing section, and `pyproject.toml` project URL (#269).
 
+
+### Changed
+
+- **`creative/bg_remover`**: Hardened the skill by reusing rembg sessions across executions, validating Base64 input and image files, rejecting empty, oversized (>25 MB), or invalid images, and automatically creating parent directories for `output_path`. Updated documentation and expanded bundle tests for the new validation behavior.
+
 ## [0.4.7] - 2026-07-25
 
 ### Added
diff --git a/docs/skills/bg_remover.md b/docs/skills/bg_remover.md
index 36823a3..3aed6a4 100644
--- a/docs/skills/bg_remover.md
+++ b/docs/skills/bg_remover.md
@@ -74,7 +74,7 @@ Agent routing details live in `skills/creative/bg_remover/instructions.md`.
 | `height` | Output image height in pixels |
 | `model_used` | rembg model used for processing |
 | `error` | Human-readable error message when `success` is `false` |
-| `error_code` | `INVALID_INPUT`, `MISSING_DEPENDENCY`, or `PROCESSING_FAILED` |
+| `error_code` | `INVALID_INPUT`, `FILE_NOT_FOUND`, `MISSING_DEPENDENCY`, or `PROCESSING_FAILED` |
 
 ## Input scenarios
 
@@ -123,7 +123,7 @@ Skill instructions: when to invoke, input/output conventions, URL and cloud pre-
 
 ### The Body (`skill.py`)
 
-Lazy-imports `rembg` and Pillow, runs `new_session` + `remove`, and returns structured JSON with transparent PNG bytes.
+Lazy-imports `rembg` and Pillow, reuses cached rembg sessions, runs `remove`, and returns structured JSON with transparent PNG bytes.
 
 ## Usage Examples
 
@@ -133,7 +133,7 @@ Use `bundle["class"]()` in the snippets below; explicit `bundle["module"].ClassN
 
 ### Runnable examples
 
-See [examples/README.md](../../examples/README.md) for the current runnable-script inventory. There is no dedicated runnable example for this skill yet; the Claude, OpenAI, and DeepSeek sections below are **catalog snippets only** (same pattern as [PDF Form Filler](pdf_form_filler.md)).
+See [examples/README.md](../../examples/README.md) for the current runnable-script inventory. See `examples/bg_remover_demo.py` for a runnable local example. The Claude, OpenAI, and DeepSeek sections below remain catalog snippets showing provider integration patterns.
 
 Sample user request:
 
diff --git a/docs/usage/agent_loops.md b/docs/usage/agent_loops.md
index 8f699ca..2384765 100644
--- a/docs/usage/agent_loops.md
+++ b/docs/usage/agent_loops.md
@@ -98,7 +98,7 @@ skills in one harness.
 | `office/pdf_form_filler` | - | `gemini_pdf_form_filler.py` | `claude_pdf_form_filler.py` | (catalog page) | (catalog page) | `ollama_skills_test.py` (multi-skill) |
 | `compliance/mica_module` | - | `mica_rag_flow.py` | `mica_claude_flow.py` | (catalog page) | (catalog page) | `mica_ollama_flow.py` |
 | `compliance/pii_masker` | `pii_guardrail_flow.py` (local execute) | (catalog page) | (catalog page) | (catalog page) | (catalog page) | (catalog page) |
-| `creative/bg_remover` | (catalog page) | (catalog page) | (catalog page) | (catalog page) | (catalog page) | (catalog page) |
+| `creative/bg_remover` | `bg_remover_demo.py` (local execute) | (catalog page) | (catalog page) | (catalog page) | (catalog page) | (catalog page) |
 | `optimization/prompt_rewriter` | `prompt_compression_demo.py` (local execute) | (catalog page) | (catalog page) | (catalog page) | (catalog page) | `ollama_skills_test.py` (multi-skill) |
 | `data_engineering/synthetic_generator` | `build_dataset_demo.py` (local execute, Gemini backend) | (catalog page) | (catalog page) | (catalog page) | (catalog page) | (catalog page) |
 | `data_engineering/novelty_extractor` | `novelty_extractor_demo.py` (local execute) | `gemini_novelty_extractor.py` | (catalog page) | (catalog page) | (catalog page) | `ollama_novelty_extractor.py` |
diff --git a/examples/README.md b/examples/README.md
index 6bdbf38..52c39d3 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -1,80 +1,70 @@
-# Skillware Examples Index

-

-> **These are usage examples, not tests.** Runnable provider demos live here; automated tests live in `skills/**/test_skill.py` (bundle) and `tests/` (framework and optional maintainer depth). See [TESTING.md](../docs/TESTING.md).

-

-Runnable examples in this directory show how to load Skillware skills, adapt

-them for a provider, execute local skill logic, and return tool results to an

-agent loop. After `pip install skillware`, run `skillware examples` or

-`skillware list --examples` from any directory to browse the index in the

-terminal; when no local `examples/README.md` is present, the index is fetched

-from GitHub (network required); see

-[CLI reference](../docs/usage/cli.md). Provider setup details live in the usage guides:

-

-- [API keys for skills](../docs/usage/api_keys.md)

-- [Gemini](../docs/usage/gemini.md)

-- [Claude](../docs/usage/claude.md)

-- [OpenAI](../docs/usage/openai.md)

+# Skillware Examples Index
+
+> **These are usage examples, not tests.** Runnable provider demos live here; automated tests live in `skills/**/test_skill.py` (bundle) and `tests/` (framework and optional maintainer depth). See [TESTING.md](../docs/TESTING.md).
+
+Runnable examples in this directory show how to load Skillware skills, adapt them for a provider, execute local skill logic, and return tool results to an agent loop. After `pip install skillware`, run `skillware examples` or `skillware list --examples` from any directory to browse the index in the terminal; when no local `examples/README.md` is present, the index is fetched from GitHub (network required); see [CLI reference](../docs/usage/cli.md). Provider setup details live in the usage guides:
+
+- [API keys for skills](../docs/usage/api_keys.md)
+- [Gemini](../docs/usage/gemini.md)
+- [Claude](../docs/usage/claude.md)
+- [OpenAI](../docs/usage/openai.md)
 - [OpenAI-compatible hosts](../docs/usage/openai_compatible.md)
-- [DeepSeek](../docs/usage/deepseek.md)

-- [Ollama](../docs/usage/ollama.md)

-- [Install extras](../docs/usage/install_extras.md)

-- [Agent loops](../docs/usage/agent_loops.md)

-

-Install the **skill extra** for each script (see [Install extras](../docs/usage/install_extras.md)), plus an **SDK extra** when the provider column is not local execute:

-

-```bash

-pip install "skillware[office_pdf_form_filler]" "skillware[gemini]"

-```

-

-For local development with every skill and SDK:

-

-```bash

-pip install -e ".[dev,all,agents]"

-```

-

-## Runnable Scripts

-

-| Script | Skill ID | Provider | Required extra | Required env vars | Description |

-| :--- | :--- | :--- | :--- | :--- | :--- |

-| `mental_coach_demo.py` | `wellness/mental_coach` | Local execute | `[wellness_mental_coach]` | None | Demonstrates coaching, crisis escalation, and blocked clinical paths locally. |

-| `build_dataset_demo.py` | `data_engineering/synthetic_generator` | Local execute (Gemini backend) | `[data_engineering_synthetic_generator]`, `[gemini]` | `GOOGLE_API_KEY` | Generates a JSONL synthetic dataset with the synthetic generator skill. |

-| `claude_pdf_form_filler.py` | `office/pdf_form_filler` | Claude | `[office_pdf_form_filler]`, `[claude]` | `ANTHROPIC_API_KEY` | Uses Claude with the PDF form filler skill to map instructions to fields. |

-| `claude_tos_evaluator.py` | `compliance/tos_evaluator` | Claude | `[compliance_tos_evaluator]`, `[claude]` | `ANTHROPIC_API_KEY` | Runs a Claude tool loop for website automation policy review. |

-| `claude_issue_resolver.py` | `dev_tools/issue_resolver` | Claude | `[dev_tools_issue_resolver]`, `[claude]` | `ANTHROPIC_API_KEY`; optional `GITHUB_TOKEN` | Claude loop for GitHub issue analysis; fetches issue data after `prepare` (sample: issue #123). |

-| `claude_wallet_check.py` | `finance/wallet_screening` | Claude | `[finance_wallet_screening]`, `[claude]` | `ANTHROPIC_API_KEY`, `ETHERSCAN_API_KEY` | Screens an Ethereum wallet and returns the result through a Claude tool loop. |

-| `deepseek_tos_evaluator.py` | `compliance/tos_evaluator` | DeepSeek | `[compliance_tos_evaluator]`, `[openai]` | `DEEPSEEK_API_KEY` | Uses the OpenAI-compatible DeepSeek API for terms-of-service evaluation. |

-| `gemini_pdf_form_filler.py` | `office/pdf_form_filler` | Gemini | `[office_pdf_form_filler]`, `[gemini]` | `GOOGLE_API_KEY`, `ANTHROPIC_API_KEY` | Uses Gemini as the agent while the PDF skill calls Anthropic for form filling. |

-| `gemini_tos_evaluator.py` | `compliance/tos_evaluator` | Gemini | `[compliance_tos_evaluator]`, `[gemini]` | `GOOGLE_API_KEY` | Runs the terms-of-service evaluator with a Gemini function-calling loop. |

-| `gemini_wallet_check.py` | `finance/wallet_screening` | Gemini | `[finance_wallet_screening]`, `[gemini]` | `GOOGLE_API_KEY`, `ETHERSCAN_API_KEY` | Screens an Ethereum wallet with Gemini orchestration and Etherscan data. |

-| `gemini_evm_tx_handler.py` | `defi/evm_tx_handler` | Gemini | `[defi_evm_tx_handler]`, `[gemini]` | `GOOGLE_API_KEY`; for live swaps also `AGENT_WALLET_PRIVATE_KEY`, `BASE_RPC_URL` or `ETHEREUM_RPC_URL`. Set `EVM_TX_HANDLER_EXAMPLE_DEMO=1` for mocked flow without keys. | Resolve → quote → preview → execute buy flow via Gemini tool loop (or demo mode). |

-| `claude_evm_tx_handler.py` | `defi/evm_tx_handler` | Claude | `[defi_evm_tx_handler]`, `[claude]` | `ANTHROPIC_API_KEY`; for live swaps also `AGENT_WALLET_PRIVATE_KEY`, RPC URLs. Demo: `EVM_TX_HANDLER_EXAMPLE_DEMO=1`. | Claude tool loop for structured DeFi intent and optional execute after confirmation. |

-| `mica_claude_flow.py` | `compliance/mica_module` | Claude | `[compliance_mica_module]`, `[claude]` | `ANTHROPIC_API_KEY` | Runs a MiCA compliance agent loop through Claude. |

-| `mica_ollama_flow.py` | `compliance/mica_module` | Ollama | `[compliance_mica_module]`; install `ollama` separately | None | Runs a local Ollama MiCA flow with prompt-mode tool calling. |

-| `mica_rag_flow.py` | `compliance/mica_module` | Gemini | `[compliance_mica_module]`, `[gemini]` | `GOOGLE_API_KEY` | Runs the MiCA RAG flow with Gemini. |

-| `ollama_skills_test.py` | `finance/wallet_screening`, `office/pdf_form_filler`, `optimization/prompt_rewriter` | Ollama | `[finance_wallet_screening]`, `[office_pdf_form_filler]`, `[optimization_prompt_rewriter]`; install `ollama` separately | `ETHERSCAN_API_KEY`, `ANTHROPIC_API_KEY` | Loads multiple skills and tests prompt-mode tool calling with Ollama. |

-| `ollama_tos_evaluator.py` | `compliance/tos_evaluator` | Ollama | `[compliance_tos_evaluator]`; install `ollama` separately | None | Runs the terms-of-service evaluator with local Ollama prompt-mode calls. |

-| `openai_tos_evaluator.py` | `compliance/tos_evaluator` | OpenAI | `[compliance_tos_evaluator]`, `[openai]` | `OPENAI_API_KEY` | Runs the terms-of-service evaluator with OpenAI function calling. |

+- [DeepSeek](../docs/usage/deepseek.md)
+- [Ollama](../docs/usage/ollama.md)
+- [Install extras](../docs/usage/install_extras.md)
+- [Agent loops](../docs/usage/agent_loops.md)
+
+Install the **skill extra** for each script (see [Install extras](../docs/usage/install_extras.md)), plus an **SDK extra** when the provider column is not local execute:
+
+```bash
+pip install "skillware[office_pdf_form_filler]" "skillware[gemini]"
+```
+
+For local development with every skill and SDK:
+
+```bash
+pip install -e ".[dev,all,agents]"
+```
+
+## Runnable Scripts
+
+| Script | Skill ID | Provider | Required extra | Required env vars | Description |
+| :--- | :--- | :--- | :--- | :--- | :--- |
+| `mental_coach_demo.py` | `wellness/mental_coach` | Local execute | `[wellness_mental_coach]` | None | Demonstrates coaching, crisis escalation, and blocked clinical paths locally. |
+| `build_dataset_demo.py` | `data_engineering/synthetic_generator` | Local execute (Gemini backend) | `[data_engineering_synthetic_generator]`, `[gemini]` | `GOOGLE_API_KEY` | Generates a JSONL synthetic dataset with the synthetic generator skill. |
+| `claude_pdf_form_filler.py` | `office/pdf_form_filler` | Claude | `[office_pdf_form_filler]`, `[claude]` | `ANTHROPIC_API_KEY` | Uses Claude with the PDF form filler skill to map instructions to fields. |
+| `claude_tos_evaluator.py` | `compliance/tos_evaluator` | Claude | `[compliance_tos_evaluator]`, `[claude]` | `ANTHROPIC_API_KEY` | Runs a Claude tool loop for website automation policy review. |
+| `claude_issue_resolver.py` | `dev_tools/issue_resolver` | Claude | `[dev_tools_issue_resolver]`, `[claude]` | `ANTHROPIC_API_KEY`; optional `GITHUB_TOKEN` | Claude loop for GitHub issue analysis; fetches issue data after `prepare` (sample: issue #123). |
+| `claude_wallet_check.py` | `finance/wallet_screening` | Claude | `[finance_wallet_screening]`, `[claude]` | `ANTHROPIC_API_KEY`, `ETHERSCAN_API_KEY` | Screens an Ethereum wallet and returns the result through a Claude tool loop. |
+| `deepseek_tos_evaluator.py` | `compliance/tos_evaluator` | DeepSeek | `[compliance_tos_evaluator]`, `[openai]` | `DEEPSEEK_API_KEY` | Uses the OpenAI-compatible DeepSeek API for terms-of-service evaluation. |
+| `gemini_pdf_form_filler.py` | `office/pdf_form_filler` | Gemini | `[office_pdf_form_filler]`, `[gemini]` | `GOOGLE_API_KEY`, `ANTHROPIC_API_KEY` | Uses Gemini as the agent while the PDF skill calls Anthropic for form filling. |
+| `gemini_tos_evaluator.py` | `compliance/tos_evaluator` | Gemini | `[compliance_tos_evaluator]`, `[gemini]` | `GOOGLE_API_KEY` | Runs the terms-of-service evaluator with a Gemini function-calling loop. |
+| `gemini_wallet_check.py` | `finance/wallet_screening` | Gemini | `[finance_wallet_screening]`, `[gemini]` | `GOOGLE_API_KEY`, `ETHERSCAN_API_KEY` | Screens an Ethereum wallet with Gemini orchestration and Etherscan data. |
+| `gemini_evm_tx_handler.py` | `defi/evm_tx_handler` | Gemini | `[defi_evm_tx_handler]`, `[gemini]` | `GOOGLE_API_KEY`; for live swaps also `AGENT_WALLET_PRIVATE_KEY`, `BASE_RPC_URL` or `ETHEREUM_RPC_URL`. Set `EVM_TX_HANDLER_EXAMPLE_DEMO=1` for mocked flow without keys. | Resolve → quote → preview → execute buy flow via Gemini tool loop (or demo mode). |
+| `claude_evm_tx_handler.py` | `defi/evm_tx_handler` | Claude | `[defi_evm_tx_handler]`, `[claude]` | `ANTHROPIC_API_KEY`; for live swaps also `AGENT_WALLET_PRIVATE_KEY`, RPC URLs. Demo: `EVM_TX_HANDLER_EXAMPLE_DEMO=1`. | Claude tool loop for structured DeFi intent and optional execute after confirmation. |
+| `mica_claude_flow.py` | `compliance/mica_module` | Claude | `[compliance_mica_module]`, `[claude]` | `ANTHROPIC_API_KEY` | Runs a MiCA compliance agent loop through Claude. |
+| `mica_ollama_flow.py` | `compliance/mica_module` | Ollama | `[compliance_mica_module]`; install `ollama` separately | None | Runs a local Ollama MiCA flow with prompt-mode tool calling. |
+| `mica_rag_flow.py` | `compliance/mica_module` | Gemini | `[compliance_mica_module]`, `[gemini]` | `GOOGLE_API_KEY` | Runs the MiCA RAG flow with Gemini. |
+| `ollama_skills_test.py` | `finance/wallet_screening`, `office/pdf_form_filler`, `optimization/prompt_rewriter` | Ollama | `[finance_wallet_screening]`, `[office_pdf_form_filler]`, `[optimization_prompt_rewriter]`; install `ollama` separately | `ETHERSCAN_API_KEY`, `ANTHROPIC_API_KEY` | Loads multiple skills and tests prompt-mode tool calling with Ollama. |
+| `ollama_tos_evaluator.py` | `compliance/tos_evaluator` | Ollama | `[compliance_tos_evaluator]`; install `ollama` separately | None | Runs the terms-of-service evaluator with local Ollama prompt-mode calls. |
+| `openai_tos_evaluator.py` | `compliance/tos_evaluator` | OpenAI | `[compliance_tos_evaluator]`, `[openai]` | `OPENAI_API_KEY` | Runs the terms-of-service evaluator with OpenAI function calling. |
 | `openai_compatible_host.py` | `compliance/tos_evaluator` | Groq (OpenAI-compatible) | `[compliance_tos_evaluator]`, `[openai]` | `GROQ_API_KEY`, `GROQ_MODEL` | Runs the terms-of-service evaluator through Groq's OpenAI-compatible API. |
-| `pii_guardrail_flow.py` | `compliance/pii_masker` | Local execute | `[compliance_pii_masker]` | None | Demonstrates local PII masking before passing text to an external agent. |

-| `prompt_compression_demo.py` | `optimization/prompt_rewriter` | Local execute | `[optimization_prompt_rewriter]` | None | Demonstrates prompt compression without a provider loop. |

-| `novelty_extractor_demo.py` | `data_engineering/novelty_extractor` | Local execute | `[data_engineering_novelty_extractor]` | None | Demonstrates multi-turn corpus distillation using local embeddings with no API key. |

-| `gemini_novelty_extractor.py` | `data_engineering/novelty_extractor` | Gemini | `[data_engineering_novelty_extractor]`, `[gemini]` | `GOOGLE_API_KEY` | Runs the novelty extractor with a Gemini function-calling loop. |

-| `ollama_novelty_extractor.py` | `data_engineering/novelty_extractor` | Ollama | `[data_engineering_novelty_extractor]`; install `ollama` separately | None | Runs the novelty extractor with local Ollama prompt-mode calls. |

-| `gemini_issue_resolver.py` | `dev_tools/issue_resolver` | Gemini | `[dev_tools_issue_resolver]`, `[gemini]` | `GOOGLE_API_KEY`; optional `GITHUB_TOKEN` | Gemini loop for GitHub issue analysis; fetches issue data after `prepare` (sample: issue #123). |

-| `ollama_issue_resolver.py` | `dev_tools/issue_resolver` | Ollama | `[dev_tools_issue_resolver]`; install `ollama` separately | optional `GITHUB_TOKEN`; `OLLAMA_MODEL` (default `gemma4:e2b`) | Ollama prompt-mode loop for GitHub issue analysis (sample: issue #123). |

-| `token_limiter_loop.py` | `monitoring/token_limiter` | Local execute | `[monitoring_token_limiter]` | None | Simulates a runaway task hitting a token ceiling with deterministic budget checks. |

-| `gemini_token_limiter.py` | `monitoring/token_limiter` | Gemini | `[monitoring_token_limiter]`, `[gemini]` | Optional `GOOGLE_API_KEY` for Phase 2 live loop | Local budget simulation plus optional Gemini tool loop. |

-| `claude_token_limiter.py` | `monitoring/token_limiter` | Claude | `[monitoring_token_limiter]`, `[claude]` | Optional `ANTHROPIC_API_KEY` for Phase 2 live loop | Local budget simulation plus optional Claude tool loop. |

-| `gemini_uk_companies_house_handler.py` | `finance/uk_companies_house_handler` | Gemini | `[finance_uk_companies_house_handler]`, `[gemini]` | `GOOGLE_API_KEY`, `COMPANIES_HOUSE_API_KEY` | Resolve company, officers, filings via Gemini tool loop. |

-| `uk_companies_house_handler_demo.py` | `finance/uk_companies_house_handler` | Local execute | `[finance_uk_companies_house_handler]` | None | Runs a scripted sequence (resolve, profile, officers, pscs, filings) with mocked HTTP responses (no API keys needed). |

-

-## Notes

-

-- **Skill extras** are listed per script above. Empty extras (`[]` in `pyproject.toml`) still install core + skillware; use them so docs stay stable when a skill gains new runtime deps.

-- Agent-side model keys such as `GOOGLE_API_KEY`, `ANTHROPIC_API_KEY`,

-  `OPENAI_API_KEY`, and `DEEPSEEK_API_KEY` are documented in the provider

-  guides.

-- Skill runtime keys such as `ETHERSCAN_API_KEY` are documented in each skill

-  manifest and on the skill catalog pages.

-- Ollama examples require the Python `ollama` package, a local Ollama server,

-  and the model named in the script, but no cloud API key.

+| `pii_guardrail_flow.py` | `compliance/pii_masker` | Local execute | `[compliance_pii_masker]` | None | Demonstrates local PII masking before passing text to an external agent. |
+| `prompt_compression_demo.py` | `optimization/prompt_rewriter` | Local execute | `[optimization_prompt_rewriter]` | None | Demonstrates prompt compression without a provider loop. |
+| `novelty_extractor_demo.py` | `data_engineering/novelty_extractor` | Local execute | `[data_engineering_novelty_extractor]` | None | Demonstrates multi-turn corpus distillation using local embeddings with no API key. |
+| `gemini_novelty_extractor.py` | `data_engineering/novelty_extractor` | Gemini | `[data_engineering_novelty_extractor]`, `[gemini]` | `GOOGLE_API_KEY` | Runs the novelty extractor with a Gemini function-calling loop. |
+| `ollama_novelty_extractor.py` | `data_engineering/novelty_extractor` | Ollama | `[data_engineering_novelty_extractor]`; install `ollama` separately | None | Runs the novelty extractor with local Ollama prompt-mode calls. |
+| `gemini_issue_resolver.py` | `dev_tools/issue_resolver` | Gemini | `[dev_tools_issue_resolver]`, `[gemini]` | `GOOGLE_API_KEY`; optional `GITHUB_TOKEN` | Gemini loop for GitHub issue analysis; fetches issue data after `prepare` (sample: issue #123). |
+| `ollama_issue_resolver.py` | `dev_tools/issue_resolver` | Ollama | `[dev_tools_issue_resolver]`; install `ollama` separately | optional `GITHUB_TOKEN`; `OLLAMA_MODEL` (default `gemma4:e2b`) | Ollama prompt-mode loop for GitHub issue analysis (sample: issue #123). |
+| `token_limiter_loop.py` | `monitoring/token_limiter` | Local execute | `[monitoring_token_limiter]` | None | Simulates a runaway task hitting a token ceiling with deterministic budget checks. |
+| `gemini_token_limiter.py` | `monitoring/token_limiter` | Gemini | `[monitoring_token_limiter]`, `[gemini]` | Optional `GOOGLE_API_KEY` for Phase 2 live loop | Local budget simulation plus optional Gemini tool loop. |
+| `claude_token_limiter.py` | `monitoring/token_limiter` | Claude | `[monitoring_token_limiter]`, `[claude]` | Optional `ANTHROPIC_API_KEY` for Phase 2 live loop | Local budget simulation plus optional Claude tool loop. |
+| `gemini_uk_companies_house_handler.py` | `finance/uk_companies_house_handler` | Gemini | `[finance_uk_companies_house_handler]`, `[gemini]` | `GOOGLE_API_KEY`, `COMPANIES_HOUSE_API_KEY` | Resolve company, officers, filings via Gemini tool loop. |
+| `uk_companies_house_handler_demo.py` | `finance/uk_companies_house_handler` | Local execute | `[finance_uk_companies_house_handler]` | None | Runs a scripted sequence (resolve, profile, officers, pscs, filings) with mocked HTTP responses (no API keys needed). |
+| `bg_remover_demo.py` | `creative/bg_remover` | Local execute | `[creative_bg_remover]` | None | Demonstrates offline background removal from a local image and optionally writes a transparent PNG. |
+## Notes
+
+- **Skill extras** are listed per script above. Empty extras (`[]` in `pyproject.toml`) still install core + skillware; use them so docs stay stable when a skill gains new runtime deps.
+- Agent-side model keys such as `GOOGLE_API_KEY`, `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, and `DEEPSEEK_API_KEY` are documented in the provider guides.
+- Skill runtime keys such as `ETHERSCAN_API_KEY` are documented in each skill manifest and on the skill catalog pages.
+- Ollama examples require the Python `ollama` package, a local Ollama server, and the model named in the script, but no cloud API key.
diff --git a/examples/bg_remover_demo.py b/examples/bg_remover_demo.py
new file mode 100644
index 0000000..e2d9901
--- /dev/null
+++ b/examples/bg_remover_demo.py
@@ -0,0 +1,41 @@
+from pathlib import Path
+
+from skillware.core.loader import SkillLoader
+
+
+def run_demo():
+    print("Loading Background Remover...")
+
+    skill_bundle = SkillLoader.load_skill("creative/bg_remover")
+    skill_instance = skill_bundle["class"]()
+
+    input_path = "examples/sample_input.png"
+    output_path = "examples/sample_output_no_bg.png"
+
+    if not Path(input_path).exists():
+        print(f"Input image not found: {input_path}")
+        print("Place a sample PNG at the above path before running this demo.")
+        return
+
+    print(f"Removing background from: {input_path}")
+
+    result = skill_instance.execute(
+        {
+            "input_path": input_path,
+            "output_path": output_path,
+        }
+    )
+
+    if result["success"]:
+        print("Background removed successfully!")
+        print(f"Saved to: {result['output_path']}")
+        print(f"Output size: {result['width']} x {result['height']}")
+        print(f"Model: {result['model_used']}")
+        print(f"Base64 length: {len(result['image_base64'])}")
+    else:
+        print(f"Failed: {result['error']}")
+        print(f"Error code: {result['error_code']}")
+
+
+if __name__ == "__main__":
+    run_demo()
diff --git a/skills/creative/bg_remover/instructions.md b/skills/creative/bg_remover/instructions.md
index ee0e3c6..1846732 100644
--- a/skills/creative/bg_remover/instructions.md
+++ b/skills/creative/bg_remover/instructions.md
@@ -32,6 +32,8 @@ For S3, GCS, Azure Blob, or Cloudflare R2: download the object to a temp file 
 
 On the **first run** in a fresh environment, processing may take a few minutes while rembg downloads the ONNX model (~176 MB for `isnet-general-use`). Later runs reuse the cache and are much faster.
 
+Background removal sessions are also reused across executions for the same model, reducing repeated initialization overhead.
+
 ## Input (one required: `image` OR `input_path`)
 
 | Scenario | Parameter |
@@ -43,6 +45,15 @@ On the **first run** in a fresh environment, processing may take a few minutes w
 
 If **both** `image` and `input_path` are sent, **`image` wins** (do not double-submit).
 
+### Input validation
+
+- Invalid Base64 payloads are rejected.
+- `input_path` must reference an existing file (directories are rejected).
+- Empty image files are rejected.
+- Images larger than **25 MB** are rejected.
+- Image integrity is validated before processing.
+- output_path must not contain path traversal components (for example `..`).
+
 ### Example payloads
 
 Chat / attachment:
@@ -91,6 +102,8 @@ Omit `model` for default `isnet-general-use`. Set `alpha_matting` only when edge
 | Chat or API only | Omit `output_path`; use `image_base64` from the result (always present on success). |
 | Save next to original | Same directory, new name (e.g. `1223_no_bg.png`). |
 
+If the parent directory of `output_path` does not exist, it is created automatically before writing the PNG.
+
 ## Interpreting a successful result
 
 When `success` is `true`:
@@ -109,10 +122,13 @@ Runtime: `rembg`, `pillow`, `onnxruntime`. Install: `pip install "skillware[crea
 
 First `execute()` downloads the ONNX model to the rembg cache (`~/.u2net/` on Linux/macOS, `%USERPROFILE%\.u2net\` on Windows).
 
+Subsequent executions reuse cached rembg sessions for the selected model, reducing repeated initialization overhead.
+
 ## Errors
 
 | `error_code` | Response |
 | :--- | :--- |
-| `INVALID_INPUT` | Ask for an upload, attachment, base64 payload, or local `input_path`. |
+| `INVALID_INPUT` | Invalid Base64, missing input, directory path, empty image, oversized image (>25 MB), corrupt image, or unsupported input. |
 | `MISSING_DEPENDENCY` | Ask the user to run `pip install "skillware[creative_bg_remover]"` (or `pip install rembg pillow onnxruntime`), then retry. |
 | `PROCESSING_FAILED` | Surface the `error` string; input may be missing, corrupt, or unsupported. |
+| `FILE_NOT_FOUND` | The supplied `input_path` does not exist. Ask the user to verify the path and try again. |
diff --git a/skills/creative/bg_remover/manifest.yaml b/skills/creative/bg_remover/manifest.yaml
index 3e59e81..c4e61c9 100644
--- a/skills/creative/bg_remover/manifest.yaml
+++ b/skills/creative/bg_remover/manifest.yaml
@@ -1,5 +1,5 @@
 name: "creative/bg_remover"
-version: "0.1.0"
+version: "0.2.0"
 description: "Remove image backgrounds locally using rembg."
 short_description: "Offline background removal using rembg with transparent PNG output."
 
@@ -77,7 +77,7 @@ outputs:
 
   error_code:
     type: string
-    description: "INVALID_INPUT, MISSING_DEPENDENCY, or PROCESSING_FAILED."
+    description: "INVALID_INPUT, FILE_NOT_FOUND, MISSING_DEPENDENCY, or PROCESSING_FAILED."
 
 requirements:
   - rembg>=2.0.0
diff --git a/skills/creative/bg_remover/skill.py b/skills/creative/bg_remover/skill.py
index ceac48e..953ca96 100644
--- a/skills/creative/bg_remover/skill.py
+++ b/skills/creative/bg_remover/skill.py
@@ -5,15 +5,38 @@
 
 from skillware.core.base_skill import BaseSkill
 
+MAX_IMAGE_BYTES = 25 * 1024 * 1024  # 25 MB
+
 
 class BackgroundRemover(BaseSkill):
     """Remove image backgrounds locally using rembg."""
 
+    _sessions = {}
+
+    @classmethod
+    def _get_session(cls, model: str):
+        """Load and reuse rembg sessions across executions."""
+        if model not in cls._sessions:
+            from rembg import new_session
+
+            cls._sessions[model] = new_session(model)
+
+        return cls._sessions[model]
+
+    @staticmethod
+    def _validate_output_path(output_path: str) -> Path:
+        output = Path(output_path)
+
+        if ".." in output.parts:
+            raise ValueError("Unsafe output_path contains path traversal.")
+
+        return output
+
     @property
     def manifest(self) -> Dict[str, Any]:
         return {
             "name": "creative/bg_remover",
-            "version": "0.1.0",
+            "version": "0.2.0",
             "description": (
                 "Remove image backgrounds locally using rembg "
                 "and return a transparent PNG."
@@ -26,7 +49,7 @@ def execute(self, params: Dict[str, Any]) -> Dict[str, Any]:
         try:
             try:
                 from PIL import Image
-                from rembg import new_session, remove
+                from rembg import remove
             except ImportError:
                 return {
                     "success": False,
@@ -53,11 +76,61 @@ def execute(self, params: Dict[str, Any]) -> Dict[str, Any]:
 
             # Read image bytes
             if image_b64:
-                image_bytes = base64.b64decode(image_b64)
+                try:
+                    image_bytes = base64.b64decode(image_b64, validate=True)
+                except Exception:
+                    return {
+                        "success": False,
+                        "error": "Invalid base64 image.",
+                        "error_code": "INVALID_INPUT",
+                    }
             else:
-                image_bytes = Path(input_path).read_bytes()
+                input_file = Path(input_path)
+
+                if input_file.is_dir():
+                    return {
+                        "success": False,
+                        "error": "Input path must be a file, not a directory.",
+                        "error_code": "INVALID_INPUT",
+                    }
+
+                if not input_file.exists():
+                    return {
+                        "success": False,
+                        "error": f"Input file '{input_path}' was not found.",
+                        "error_code": "FILE_NOT_FOUND",
+                    }
+
+                image_bytes = input_file.read_bytes()
+
+            if len(image_bytes) > MAX_IMAGE_BYTES:
+                return {
+                    "success": False,
+                    "error": f"Input image exceeds the maximum size of {MAX_IMAGE_BYTES // (1024 * 1024)} MB.",
+                    "error_code": "INVALID_INPUT",
+                }
+
+            if len(image_bytes) == 0:
+                return {
+                    "success": False,
+                    "error": "Input image is empty.",
+                    "error_code": "INVALID_INPUT",
+                }
+
+            try:
+                image = Image.open(io.BytesIO(image_bytes))
+                image.verify()
+
+                # Reopen after verify() because verify() leaves the image unusable.
+                image = Image.open(io.BytesIO(image_bytes))
+            except Exception:
+                return {
+                    "success": False,
+                    "error": "Input is not a valid image.",
+                    "error_code": "INVALID_INPUT",
+                }
 
-            session = new_session(model)
+            session = self._get_session(model)
 
             output_bytes = remove(
                 image_bytes,
@@ -76,7 +149,17 @@ def execute(self, params: Dict[str, Any]) -> Dict[str, Any]:
 
             # Optional save
             if output_path:
-                Path(output_path).write_bytes(buffer.getvalue())
+                try:
+                    output_file = self._validate_output_path(output_path)
+                except ValueError:
+                    return {
+                        "success": False,
+                        "error": "Unsafe output_path.",
+                        "error_code": "INVALID_INPUT",
+                    }
+
+                output_file.parent.mkdir(parents=True, exist_ok=True)
+                output_file.write_bytes(buffer.getvalue())
 
             return {
                 "success": True,
diff --git a/skills/creative/bg_remover/test_skill.py b/skills/creative/bg_remover/test_skill.py
index 92f50fb..9d141e2 100644
--- a/skills/creative/bg_remover/test_skill.py
+++ b/skills/creative/bg_remover/test_skill.py
@@ -9,7 +9,7 @@
 
 from skillware.core.loader import SkillLoader
 
-from .skill import BackgroundRemover
+from .skill import BackgroundRemover, MAX_IMAGE_BYTES
 
 import sys
 import types
@@ -117,7 +117,7 @@ def test_invalid_base64(skill):
     result = skill.execute({"image": "not_base64"})
 
     assert result["success"] is False
-    assert result["error_code"] == "PROCESSING_FAILED"
+    assert result["error_code"] == "INVALID_INPUT"
 
 
 def test_missing_dependency(monkeypatch, skill):
@@ -249,4 +249,101 @@ def test_missing_input_path(skill):
     result = skill.execute({"input_path": "/nonexistent/path/image.png"})
 
     assert result["success"] is False
-    assert result["error_code"] == "PROCESSING_FAILED"
+    assert result["error_code"] == "FILE_NOT_FOUND"
+
+
+def test_directory_input(skill, tmp_path):
+    result = skill.execute({"input_path": str(tmp_path)})
+
+    assert result["success"] is False
+    assert result["error_code"] == "INVALID_INPUT"
+
+
+def test_empty_input_file(skill, tmp_path):
+    empty = tmp_path / "empty.png"
+    empty.write_bytes(b"")
+
+    result = skill.execute({"input_path": str(empty)})
+
+    assert result["success"] is False
+    assert result["error_code"] == "INVALID_INPUT"
+
+
+def test_invalid_image_file(skill, tmp_path):
+    bad = tmp_path / "bad.png"
+    bad.write_text("not an image")
+
+    result = skill.execute({"input_path": str(bad)})
+
+    assert result["success"] is False
+    assert result["error_code"] == "INVALID_INPUT"
+
+
+def test_output_directory_created(skill, tmp_path):
+    image = Image.new("RGB", (64, 64), "white")
+
+    input_file = tmp_path / "input.png"
+    image.save(input_file)
+
+    output_file = tmp_path / "nested" / "folder" / "output.png"
+
+    result = skill.execute(
+        {
+            "input_path": str(input_file),
+            "output_path": str(output_file),
+        }
+    )
+
+    assert result["success"] is True
+    assert output_file.exists()
+
+
+def test_session_reuse(skill, monkeypatch):
+    calls = {"count": 0}
+
+    def fake_remove(image_bytes, *args, **kwargs):
+        img = Image.new("RGBA", (100, 100), (255, 0, 0, 0))
+        buffer = io.BytesIO()
+        img.save(buffer, format="PNG")
+        return buffer.getvalue()
+
+    def fake_new_session(model_name="u2net", *args, **kwargs):
+        calls["count"] += 1
+        return object()
+
+    fake_module = types.ModuleType("rembg")
+    fake_module.remove = fake_remove
+    fake_module.new_session = fake_new_session
+    fake_module.__spec__ = importlib.util.spec_from_loader("rembg", loader=None)
+
+    monkeypatch.setitem(sys.modules, "rembg", fake_module)
+
+    BackgroundRemover._sessions.clear()
+
+    skill.execute({"image": create_image()})
+    skill.execute({"image": create_image()})
+
+    assert calls["count"] == 1
+
+    BackgroundRemover._sessions.clear()
+
+
+def test_unsafe_output_path(skill):
+    result = skill.execute(
+        {
+            "image": create_image(),
+            "output_path": "../../../evil.png",
+        }
+    )
+
+    assert result["success"] is False
+    assert result["error_code"] == "INVALID_INPUT"
+
+
+def test_oversized_base64_image(skill):
+    oversized = base64.b64encode(b"\0" * (MAX_IMAGE_BYTES + 1)).decode()
+
+    result = skill.execute({"image": oversized})
+
+    assert result["success"] is False
+    assert result["error_code"] == "INVALID_INPUT"