agent-skillforge 0.2.0__tar.gz → 0.3.0__tar.gz

{agent_skillforge-0.2.0/src/agent_skillforge.egg-info → agent_skillforge-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-skillforge
-Version: 0.2.0
+Version: 0.3.0
 Summary: Universal package manager, security auditor, and MCP server for portable agent skills (SKILL.md).
 Author-email: gintingism <gintingism@users.noreply.github.com>
 License-Expression: MIT
@@ -75,11 +75,16 @@ skillforge add anthropics/skills:pdf
 skillforge add gintingism/skillforge:systematic-debugging
 ```
 
-SkillForge fetches the skill, runs a security check against it, installs it into your project's active assistant folders, and pins it in `skillforge.lock`.
+SkillForge fetches the skill, runs a security check against it, installs it into your project's active assistant folders, and pins its SHA-256 package digest in `skillforge.lock`.
+
+To install and verify all locked skills across a team checkout or CI environment:
+```bash
+skillforge sync
+```
 
 ### 2. Audit a skill before running it (`audit`)
 
-Before letting an agent read and execute a skill, inspect it for hidden payloads and prompt injection:
+Before letting an agent read and execute a skill, inspect it for hidden payloads, broken progressive disclosure links, and prompt injection:
 
 ```bash
 skillforge audit ./skills/my-new-skill
@@ -89,6 +94,7 @@ Example output:
 ```text
 Audit Report: my-new-skill
 Status: PASSED (Score: 100/100)
+Context Load: 24 pointer tokens | 280 body tokens (42 lines)
 Total Findings: 0
 ```
 
@@ -97,14 +103,33 @@ What the auditor scans for:
 - **Prompt Injections & Jailbreaks (SEC-002):** Directives like `"ignore previous instructions"`, `"system override"`, or attempts to disable safety guardrails.
 - **Dangerous Shell Commands (SEC-003):** Script patterns like `curl ... | bash`, base64-decoded execution, and outbound data exfiltration.
 - **Credential Path Access (SEC-004):** Unsanctioned attempts to access `~/.ssh`, `~/.aws/credentials`, or local environment secrets.
-- **Trigger Quality (QUAL-001):** Checks whether descriptions contain explicit activation conditions for progressive disclosure.
+- **Trigger Quality (QUAL-001):** Checks whether model-invoked descriptions contain explicit activation conditions (skipped when `disable-model-invocation: true`).
+- **Directory Alignment (QUAL-002):** Verifies the package folder name matches the skill `name` in YAML frontmatter.
+- **Progressive Disclosure Budget (QUAL-003):** Flags `SKILL.md` bodies exceeding 500 lines or ~5,000 tokens so heavy reference docs are moved into `references/`.
+- **Broken Disclosure Pointers (QUAL-004):** Verifies every `references/...` or `scripts/...` file cited in `SKILL.md` actually exists in the package.
 
 For CI pipelines, pass `--strict` to fail on any warning, or `--json` for automation:
 ```bash
 skillforge audit ./skills/my-new-skill --strict --json
 ```
 
-### 3. Check what's installed in your project (`list` & `remove`)
+### 3. Profile your project's context token budget (`budget`)
+
+Every model-invoked skill keeps its `name` and `description` in the startup context window on every turn. Measure how much context your skills consume before they even trigger:
+
+```bash
+skillforge budget
+# Or inspect the built-in catalog:
+skillforge budget --builtin
+```
+```text
+Context Budget Profile (19 skills)
+  Always-Loaded Startup Pointer Cost: ~912 tokens
+  On-Demand SKILL.md Body Cost:       ~5420 tokens
+  Disclosed Resources Cost:           ~245 tokens
+```
+
+### 4. Check what's installed in your project (`list` & `remove`)
 
 See which skills are active across your editor targets:
 
@@ -126,7 +151,7 @@ To remove a skill from all detected assistants and update your lockfile:
 skillforge remove antislop-code
 ```
 
-### 4. Create and package your own skills (`create` & `pack`)
+### 5. Create and package your own skills (`create` & `pack`)
 
 Scaffold a single skill file or a full directory package with scripts and reference docs:
 
@@ -169,10 +194,12 @@ skillforge install systematic-debugging --global
 
 ## Shipped Built-in Skills
 
-SkillForge ships with 16 practical engineering skills designed to counteract common agent failure modes identified in benchmark research (SWE-bench):
+SkillForge ships with 19 practical engineering skills designed to counteract common agent failure modes identified in benchmark research (SWE-bench):
 
-### Code Quality & Architectural Boundaries
+### Code Quality, Security & Performance
 - `blast-radius-containment`: Upstream reference tracing and a strict "file touch budget" before modifying shared code or public APIs.
+- `security-threat-modeler`: Trust-boundary mapping, IDOR/BOLA object-level authorization checks, SSRF validation, and injection sink verification.
+- `performance-profiler`: Baseline-first latency (`p50`/`p95`/`p99`), N+1 query detection, and algorithmic complexity verification before optimizing.
 - `antislop-code`: Comment hygiene and anti-overengineering filters (removes trivial narrative comments, stops speculative generality).
 - `spec-driven-development`: Schema-first API design and red-green verification loops before writing application logic.
 - `zero-downtime-migration`: Four-phase Expand-Contract pattern for non-breaking database schema and API payload evolutions.
@@ -184,6 +211,7 @@ SkillForge ships with 16 practical engineering skills designed to counteract com
 - `runtime-api-verification`: Interactive signature introspection (`inspect`, `dir()`, compiler definitions) to prevent hallucinated kwargs.
 
 ### Context Efficiency & Multi-Agent Operations
+- `skill-architect`: Multi-file progressive disclosure guide (bundled with `references/mechanics.md`) for authoring low-token `SKILL.md`, `AGENTS.md`, and `CLAUDE.md` files.
 - `caveman`: Terse, high-density technical communication mode that reduces token spend while preserving exact paths, commands, and errors.
 - `context-compactor`: Structured state distillation and active hypothesis ledgers for long-running agent threads nearing context limits.
 - `mcp-tool-designer`: Type-safe schema design, idempotent hints, token boundaries, and stderr error separation for MCP servers.
@@ -216,6 +244,7 @@ SkillForge can run as an MCP server so your agent can search, read, and audit sk
 Exposed MCP tools:
 - `search_skills(query, limit)`: Search local and built-in skill catalogs.
 - `get_skill(name)`: Read the full `SKILL.md` instructions.
+- `get_skill_resource(name, resource_path)`: Read disclosed `references/*` or `scripts/*` files from a skill package on demand.
 - `validate_skill(content)`: Test frontmatter syntax.
 - `audit_skill(content)`: Audit a raw skill document for prompt injections and malicious patterns.
 - `skill://{name}`: Resource URI for direct document retrieval.

{agent_skillforge-0.2.0 → agent_skillforge-0.3.0}/README.md

@@ -37,11 +37,16 @@ skillforge add anthropics/skills:pdf
 skillforge add gintingism/skillforge:systematic-debugging
 ```
 
-SkillForge fetches the skill, runs a security check against it, installs it into your project's active assistant folders, and pins it in `skillforge.lock`.
+SkillForge fetches the skill, runs a security check against it, installs it into your project's active assistant folders, and pins its SHA-256 package digest in `skillforge.lock`.
+
+To install and verify all locked skills across a team checkout or CI environment:
+```bash
+skillforge sync
+```
 
 ### 2. Audit a skill before running it (`audit`)
 
-Before letting an agent read and execute a skill, inspect it for hidden payloads and prompt injection:
+Before letting an agent read and execute a skill, inspect it for hidden payloads, broken progressive disclosure links, and prompt injection:
 
 ```bash
 skillforge audit ./skills/my-new-skill
@@ -51,6 +56,7 @@ Example output:
 ```text
 Audit Report: my-new-skill
 Status: PASSED (Score: 100/100)
+Context Load: 24 pointer tokens | 280 body tokens (42 lines)
 Total Findings: 0
 ```
 
@@ -59,14 +65,33 @@ What the auditor scans for:
 - **Prompt Injections & Jailbreaks (SEC-002):** Directives like `"ignore previous instructions"`, `"system override"`, or attempts to disable safety guardrails.
 - **Dangerous Shell Commands (SEC-003):** Script patterns like `curl ... | bash`, base64-decoded execution, and outbound data exfiltration.
 - **Credential Path Access (SEC-004):** Unsanctioned attempts to access `~/.ssh`, `~/.aws/credentials`, or local environment secrets.
-- **Trigger Quality (QUAL-001):** Checks whether descriptions contain explicit activation conditions for progressive disclosure.
+- **Trigger Quality (QUAL-001):** Checks whether model-invoked descriptions contain explicit activation conditions (skipped when `disable-model-invocation: true`).
+- **Directory Alignment (QUAL-002):** Verifies the package folder name matches the skill `name` in YAML frontmatter.
+- **Progressive Disclosure Budget (QUAL-003):** Flags `SKILL.md` bodies exceeding 500 lines or ~5,000 tokens so heavy reference docs are moved into `references/`.
+- **Broken Disclosure Pointers (QUAL-004):** Verifies every `references/...` or `scripts/...` file cited in `SKILL.md` actually exists in the package.
 
 For CI pipelines, pass `--strict` to fail on any warning, or `--json` for automation:
 ```bash
 skillforge audit ./skills/my-new-skill --strict --json
 ```
 
-### 3. Check what's installed in your project (`list` & `remove`)
+### 3. Profile your project's context token budget (`budget`)
+
+Every model-invoked skill keeps its `name` and `description` in the startup context window on every turn. Measure how much context your skills consume before they even trigger:
+
+```bash
+skillforge budget
+# Or inspect the built-in catalog:
+skillforge budget --builtin
+```
+```text
+Context Budget Profile (19 skills)
+  Always-Loaded Startup Pointer Cost: ~912 tokens
+  On-Demand SKILL.md Body Cost:       ~5420 tokens
+  Disclosed Resources Cost:           ~245 tokens
+```
+
+### 4. Check what's installed in your project (`list` & `remove`)
 
 See which skills are active across your editor targets:
 
@@ -88,7 +113,7 @@ To remove a skill from all detected assistants and update your lockfile:
 skillforge remove antislop-code
 ```
 
-### 4. Create and package your own skills (`create` & `pack`)
+### 5. Create and package your own skills (`create` & `pack`)
 
 Scaffold a single skill file or a full directory package with scripts and reference docs:
 
@@ -131,10 +156,12 @@ skillforge install systematic-debugging --global
 
 ## Shipped Built-in Skills
 
-SkillForge ships with 16 practical engineering skills designed to counteract common agent failure modes identified in benchmark research (SWE-bench):
+SkillForge ships with 19 practical engineering skills designed to counteract common agent failure modes identified in benchmark research (SWE-bench):
 
-### Code Quality & Architectural Boundaries
+### Code Quality, Security & Performance
 - `blast-radius-containment`: Upstream reference tracing and a strict "file touch budget" before modifying shared code or public APIs.
+- `security-threat-modeler`: Trust-boundary mapping, IDOR/BOLA object-level authorization checks, SSRF validation, and injection sink verification.
+- `performance-profiler`: Baseline-first latency (`p50`/`p95`/`p99`), N+1 query detection, and algorithmic complexity verification before optimizing.
 - `antislop-code`: Comment hygiene and anti-overengineering filters (removes trivial narrative comments, stops speculative generality).
 - `spec-driven-development`: Schema-first API design and red-green verification loops before writing application logic.
 - `zero-downtime-migration`: Four-phase Expand-Contract pattern for non-breaking database schema and API payload evolutions.
@@ -146,6 +173,7 @@ SkillForge ships with 16 practical engineering skills designed to counteract com
 - `runtime-api-verification`: Interactive signature introspection (`inspect`, `dir()`, compiler definitions) to prevent hallucinated kwargs.
 
 ### Context Efficiency & Multi-Agent Operations
+- `skill-architect`: Multi-file progressive disclosure guide (bundled with `references/mechanics.md`) for authoring low-token `SKILL.md`, `AGENTS.md`, and `CLAUDE.md` files.
 - `caveman`: Terse, high-density technical communication mode that reduces token spend while preserving exact paths, commands, and errors.
 - `context-compactor`: Structured state distillation and active hypothesis ledgers for long-running agent threads nearing context limits.
 - `mcp-tool-designer`: Type-safe schema design, idempotent hints, token boundaries, and stderr error separation for MCP servers.
@@ -178,6 +206,7 @@ SkillForge can run as an MCP server so your agent can search, read, and audit sk
 Exposed MCP tools:
 - `search_skills(query, limit)`: Search local and built-in skill catalogs.
 - `get_skill(name)`: Read the full `SKILL.md` instructions.
+- `get_skill_resource(name, resource_path)`: Read disclosed `references/*` or `scripts/*` files from a skill package on demand.
 - `validate_skill(content)`: Test frontmatter syntax.
 - `audit_skill(content)`: Audit a raw skill document for prompt injections and malicious patterns.
 - `skill://{name}`: Resource URI for direct document retrieval.

{agent_skillforge-0.2.0 → agent_skillforge-0.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "agent-skillforge"
-version = "0.2.0"
+version = "0.3.0"
 description = "Universal package manager, security auditor, and MCP server for portable agent skills (SKILL.md)."
 readme = "README.md"
 license = "MIT"

{agent_skillforge-0.2.0 → agent_skillforge-0.3.0/src/agent_skillforge.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-skillforge
-Version: 0.2.0
+Version: 0.3.0
 Summary: Universal package manager, security auditor, and MCP server for portable agent skills (SKILL.md).
 Author-email: gintingism <gintingism@users.noreply.github.com>
 License-Expression: MIT
@@ -75,11 +75,16 @@ skillforge add anthropics/skills:pdf
 skillforge add gintingism/skillforge:systematic-debugging
 ```
 
-SkillForge fetches the skill, runs a security check against it, installs it into your project's active assistant folders, and pins it in `skillforge.lock`.
+SkillForge fetches the skill, runs a security check against it, installs it into your project's active assistant folders, and pins its SHA-256 package digest in `skillforge.lock`.
+
+To install and verify all locked skills across a team checkout or CI environment:
+```bash
+skillforge sync
+```
 
 ### 2. Audit a skill before running it (`audit`)
 
-Before letting an agent read and execute a skill, inspect it for hidden payloads and prompt injection:
+Before letting an agent read and execute a skill, inspect it for hidden payloads, broken progressive disclosure links, and prompt injection:
 
 ```bash
 skillforge audit ./skills/my-new-skill
@@ -89,6 +94,7 @@ Example output:
 ```text
 Audit Report: my-new-skill
 Status: PASSED (Score: 100/100)
+Context Load: 24 pointer tokens | 280 body tokens (42 lines)
 Total Findings: 0
 ```
 
@@ -97,14 +103,33 @@ What the auditor scans for:
 - **Prompt Injections & Jailbreaks (SEC-002):** Directives like `"ignore previous instructions"`, `"system override"`, or attempts to disable safety guardrails.
 - **Dangerous Shell Commands (SEC-003):** Script patterns like `curl ... | bash`, base64-decoded execution, and outbound data exfiltration.
 - **Credential Path Access (SEC-004):** Unsanctioned attempts to access `~/.ssh`, `~/.aws/credentials`, or local environment secrets.
-- **Trigger Quality (QUAL-001):** Checks whether descriptions contain explicit activation conditions for progressive disclosure.
+- **Trigger Quality (QUAL-001):** Checks whether model-invoked descriptions contain explicit activation conditions (skipped when `disable-model-invocation: true`).
+- **Directory Alignment (QUAL-002):** Verifies the package folder name matches the skill `name` in YAML frontmatter.
+- **Progressive Disclosure Budget (QUAL-003):** Flags `SKILL.md` bodies exceeding 500 lines or ~5,000 tokens so heavy reference docs are moved into `references/`.
+- **Broken Disclosure Pointers (QUAL-004):** Verifies every `references/...` or `scripts/...` file cited in `SKILL.md` actually exists in the package.
 
 For CI pipelines, pass `--strict` to fail on any warning, or `--json` for automation:
 ```bash
 skillforge audit ./skills/my-new-skill --strict --json
 ```
 
-### 3. Check what's installed in your project (`list` & `remove`)
+### 3. Profile your project's context token budget (`budget`)
+
+Every model-invoked skill keeps its `name` and `description` in the startup context window on every turn. Measure how much context your skills consume before they even trigger:
+
+```bash
+skillforge budget
+# Or inspect the built-in catalog:
+skillforge budget --builtin
+```
+```text
+Context Budget Profile (19 skills)
+  Always-Loaded Startup Pointer Cost: ~912 tokens
+  On-Demand SKILL.md Body Cost:       ~5420 tokens
+  Disclosed Resources Cost:           ~245 tokens
+```
+
+### 4. Check what's installed in your project (`list` & `remove`)
 
 See which skills are active across your editor targets:
 
@@ -126,7 +151,7 @@ To remove a skill from all detected assistants and update your lockfile:
 skillforge remove antislop-code
 ```
 
-### 4. Create and package your own skills (`create` & `pack`)
+### 5. Create and package your own skills (`create` & `pack`)
 
 Scaffold a single skill file or a full directory package with scripts and reference docs:
 
@@ -169,10 +194,12 @@ skillforge install systematic-debugging --global
 
 ## Shipped Built-in Skills
 
-SkillForge ships with 16 practical engineering skills designed to counteract common agent failure modes identified in benchmark research (SWE-bench):
+SkillForge ships with 19 practical engineering skills designed to counteract common agent failure modes identified in benchmark research (SWE-bench):
 
-### Code Quality & Architectural Boundaries
+### Code Quality, Security & Performance
 - `blast-radius-containment`: Upstream reference tracing and a strict "file touch budget" before modifying shared code or public APIs.
+- `security-threat-modeler`: Trust-boundary mapping, IDOR/BOLA object-level authorization checks, SSRF validation, and injection sink verification.
+- `performance-profiler`: Baseline-first latency (`p50`/`p95`/`p99`), N+1 query detection, and algorithmic complexity verification before optimizing.
 - `antislop-code`: Comment hygiene and anti-overengineering filters (removes trivial narrative comments, stops speculative generality).
 - `spec-driven-development`: Schema-first API design and red-green verification loops before writing application logic.
 - `zero-downtime-migration`: Four-phase Expand-Contract pattern for non-breaking database schema and API payload evolutions.
@@ -184,6 +211,7 @@ SkillForge ships with 16 practical engineering skills designed to counteract com
 - `runtime-api-verification`: Interactive signature introspection (`inspect`, `dir()`, compiler definitions) to prevent hallucinated kwargs.
 
 ### Context Efficiency & Multi-Agent Operations
+- `skill-architect`: Multi-file progressive disclosure guide (bundled with `references/mechanics.md`) for authoring low-token `SKILL.md`, `AGENTS.md`, and `CLAUDE.md` files.
 - `caveman`: Terse, high-density technical communication mode that reduces token spend while preserving exact paths, commands, and errors.
 - `context-compactor`: Structured state distillation and active hypothesis ledgers for long-running agent threads nearing context limits.
 - `mcp-tool-designer`: Type-safe schema design, idempotent hints, token boundaries, and stderr error separation for MCP servers.
@@ -216,6 +244,7 @@ SkillForge can run as an MCP server so your agent can search, read, and audit sk
 Exposed MCP tools:
 - `search_skills(query, limit)`: Search local and built-in skill catalogs.
 - `get_skill(name)`: Read the full `SKILL.md` instructions.
+- `get_skill_resource(name, resource_path)`: Read disclosed `references/*` or `scripts/*` files from a skill package on demand.
 - `validate_skill(content)`: Test frontmatter syntax.
 - `audit_skill(content)`: Audit a raw skill document for prompt injections and malicious patterns.
 - `skill://{name}`: Resource URI for direct document retrieval.

{agent_skillforge-0.2.0 → agent_skillforge-0.3.0}/src/agent_skillforge.egg-info/SOURCES.txt

@@ -25,8 +25,12 @@ src/skillforge/builtin_skills/context-compactor/SKILL.md
 src/skillforge/builtin_skills/flaky-test-isolation/SKILL.md
 src/skillforge/builtin_skills/human-in-the-loop/SKILL.md
 src/skillforge/builtin_skills/mcp-tool-designer/SKILL.md
+src/skillforge/builtin_skills/performance-profiler/SKILL.md
 src/skillforge/builtin_skills/runtime-api-verification/SKILL.md
+src/skillforge/builtin_skills/security-threat-modeler/SKILL.md
 src/skillforge/builtin_skills/self-improving-agent/SKILL.md
+src/skillforge/builtin_skills/skill-architect/SKILL.md
+src/skillforge/builtin_skills/skill-architect/references/mechanics.md
 src/skillforge/builtin_skills/spec-driven-development/SKILL.md
 src/skillforge/builtin_skills/summarize/SKILL.md
 src/skillforge/builtin_skills/systematic-debugging/SKILL.md

{agent_skillforge-0.2.0 → agent_skillforge-0.3.0}/src/skillforge/__init__.py

@@ -1,4 +1,3 @@
 """SkillForge: portable skills for AI coding assistants."""
 
-__version__ = "0.1.0"
-
+__version__ = "0.3.0"

agent_skillforge-0.3.0/src/skillforge/builtin_skills/performance-profiler/SKILL.md

@@ -0,0 +1,35 @@
+---
+name: performance-profiler
+description: Isolate latency bottlenecks, N+1 query patterns, memory allocations, and algorithmic complexity regressions using baseline-first measurement. Use when optimizing slow endpoints, high CPU/memory usage, or database queries.
+license: MIT
+compatibility: Universal
+metadata:
+  version: "1.0.0"
+  author: "SkillForge Team"
+---
+
+# Performance Profiler Protocol
+
+Never optimize by intuition. Measure baseline latency, query count, or allocation volume before modifying code, and prove the delta with a repeatable benchmark.
+
+## 1. Capture the Baseline
+
+1. **Reproduce with Realistic Cardinality**:
+   - Run the target path against representative dataset sizes ($N \ge 10^3$ rows or items) rather than trivial unit-test fixtures ($N = 2$).
+2. **Record Primary Metrics**:
+   - Capture wall-clock latency (`p50`, `p95`, `p99`), total SQL/RPC round-trips per request, or peak heap allocation before touching implementation code.
+
+## 2. Isolate the Dominant Bottleneck
+
+- **I/O & Query Amplification (N+1)**:
+  - Check whether loop bodies issue sequential queries, network calls, or filesystem reads. Batch into a single `WHERE id IN (...)` query, join, or bulk RPC.
+- **Algorithmic Complexity ($O(N^2)$ Traps)**:
+  - Replace repeated linear scans (`item in list`, nested filter loops, repeated string/array concatenation) with hash sets, indexed maps, or single-pass accumulators.
+- **Serialization & Payload Bloat**:
+  - Project only required columns (`SELECT id, status` instead of `SELECT *`), stream large responses, and avoid re-parsing static configurations inside hot loops.
+- **Unbounded Concurrency or Lock Contention**:
+  - Bound parallel fan-out with semaphores and keep network/disk I/O outside database transactions or mutex locks.
+
+## 3. Verification Gate
+
+- Re-run the exact baseline harness after the change and report `before -> after` metrics (latency, query count, or memory) alongside passing functional tests.

agent_skillforge-0.3.0/src/skillforge/builtin_skills/security-threat-modeler/SKILL.md

@@ -0,0 +1,37 @@
+---
+name: security-threat-modeler
+description: Map trust boundaries, attack surfaces, object-level authorization (IDOR/BOLA), SSRF vectors, and injection sinks before shipping backend or agent endpoints. Use when reviewing security, designing APIs, or auditing auth flows.
+license: MIT
+compatibility: Universal
+metadata:
+  version: "1.0.0"
+  author: "SkillForge Team"
+---
+
+# Security Threat Modeler Protocol
+
+Perform structured trust-boundary and attack-surface analysis before merging authentication, API, webhook, file-handling, or tool-execution changes.
+
+## 1. Enumerate Trust Boundaries
+
+1. **Identify Untrusted Inputs**:
+   - List every external entry point: HTTP params, headers, webhook bodies, uploaded archives, database records containing user text, and LLM/tool outputs.
+2. **Trace Principal & Tenant Scope**:
+   - Verify how the caller's identity (`user_id`, `org_id`, `role`) is bound to the session token rather than read from request parameters.
+
+## 2. High-Severity Vulnerability Checklist
+
+- **Broken Object Level Authorization (IDOR / BOLA)**:
+  - Scope every read, update, and delete query by both resource ID and authenticated tenant/owner (`WHERE id = :id AND org_id = :auth_org_id`).
+- **Server-Side Request Forgery (SSRF)**:
+  - Reject user-supplied URLs targeting loopback (`127.0.0.0/8`, `::1`), link-local metadata (`169.254.169.254`), and RFC1918 private ranges; validate resolved IP addresses immediately before connecting.
+- **Path Traversal & Archive Extraction**:
+  - Resolve candidate paths with `Path.resolve()` and assert `target.is_relative_to(base_dir)` before reading, writing, or extracting zip/tar members.
+- **Command & Query Injection**:
+  - Execute subprocesses as argument arrays with `shell=False` and bind all SQL parameters via parameterized queries.
+- **Webhook & Callback Authenticity**:
+  - Verify HMAC signatures using constant-time comparison (`hmac.compare_digest`) over the raw request body and enforce timestamp replay windows.
+
+## 3. Completion Gate
+
+- Write at least one negative security test proving an unauthorized tenant, traversed path, or forged signature is rejected with an explicit error status.

agent_skillforge-0.3.0/src/skillforge/builtin_skills/skill-architect/SKILL.md

@@ -0,0 +1,37 @@
+---
+name: skill-architect
+description: Design, author, and refactor SKILL.md packages, AGENTS.md catalogs, and CLAUDE.md instructions using progressive disclosure and context-budget discipline. Use when creating or optimizing agent skills and instruction files.
+license: MIT
+compatibility: Universal
+metadata:
+  version: "1.0.0"
+  author: "SkillForge Team"
+---
+
+# Skill Architect Protocol
+
+Author high-signal, low-token agent documents (`SKILL.md`, `AGENTS.md`, `CLAUDE.md`) that maximize execution accuracy without bloating the startup context window.
+
+## 1. Three-Tier Context Hierarchy
+
+1. **Tier 1 — Always-Loaded Pointer (`name` + `description`)**:
+   - Every model-invoked skill's `name` and `description` remain in the startup context on every turn.
+   - Front-load the capability and explicit trigger condition within the first 200 characters: *"Design, author, and refactor... Use when..."*.
+   - For user-only slash workflows that should never auto-trigger, set `disable-model-invocation: true` to drop startup token cost to zero.
+2. **Tier 2 — On-Demand Procedure (`SKILL.md` body)**:
+   - Keep `SKILL.md` under **500 lines** and **5,000 tokens**.
+   - Include only the decision tree, numbered workflow steps, and verification commands needed during execution.
+3. **Tier 3 — Disclosed Deep Reference (`references/` & `scripts/`)**:
+   - Move large schemas, edge-case tables, and deterministic helpers into `references/` or `scripts/`.
+   - Link each file with an explicit load condition (for example: read `references/mechanics.md` when choosing frontmatter fields or invocation modes).
+
+## 2. Authoring Rules
+
+- **Prefer Pointers over Copies**: Point to canonical files (`src/auth/middleware.py`) instead of pasting code blocks that go stale.
+- **Lead Every Step with the Action**: Start bullet points and numbered steps with the verb or command (`Run pytest -q`, `Inspect lockfile`, `Reject unsigned payloads`).
+- **Steer Positively**: Whenever prohibiting an anti-pattern, state the concrete alternative in the same sentence (*"Do not mock the database; run against the local SQLite fixture"*).
+- **Define Checkable Exit Gates**: Replace vague directives (*"ensure high quality"*) with executable gates (`skillforge audit <path> --strict`, `pytest`, `mypy`).
+
+## 3. Progressive Disclosure Reference
+
+- Read `references/mechanics.md` when configuring frontmatter fields, `disable-model-invocation`, or context budget thresholds.

agent_skillforge-0.3.0/src/skillforge/builtin_skills/skill-architect/references/mechanics.md

@@ -0,0 +1,27 @@
+# Skill Invocation & Context Budget Mechanics
+
+## Startup Discovery Budget
+
+Agent harnesses inject a compact catalog of all model-invoked skills at session startup.
+
+| Metric | Recommended Limit | Hard Ceiling |
+|--------|-------------------|--------------|
+| `description` length | <= 220 characters | 1024 characters |
+| Startup pointer cost per skill | 25–60 tokens | 100 tokens |
+| `SKILL.md` body length | <= 300 lines | 500 lines / 5,000 tokens |
+
+## Model-Invoked vs. User-Invoked Skills
+
+1. **Model-Invoked (Default)**:
+   - `disable-model-invocation` is omitted or `false`.
+   - Must include trigger keywords (`Use when`, `When`, `For`) in `description` so the model routes matching tasks accurately (`QUAL-001`).
+2. **User-Invoked (`disable-model-invocation: true`)**:
+   - Triggered explicitly by the user via `/skill-name`.
+   - Consumes **0 startup pointer tokens**.
+   - Do not include model trigger clauses in `description`; summarize only what the command executes for human readers.
+
+## Verification Checklist
+
+1. Run `skillforge validate <skill-dir>` to verify schema compliance.
+2. Run `skillforge audit <skill-dir> --strict` to verify zero prompt-injection patterns, intact `references/` and `scripts/` pointers (`QUAL-004`), and body budget compliance (`QUAL-003`).
+3. Run `skillforge budget -p <project>` to inspect total startup pointer tokens across all installed skills.