product-forge 0.2.0__tar.gz

product_forge-0.2.0/.agents/agents/agent-ops-coordinator.md

@@ -0,0 +1,17 @@
+---
+name: agent-ops-coordinator
+description: Coordinator for agent lifecycle and product operations — evals, publishing, watches, alerts, approvals. Use for "publish agent", "agent eval", "watch", "alert", "approval".
+rule_areas: [PF-AGENT, PF-OPS]
+skills: [pf-agent-factory, pf-governance, pf-reporting]
+---
+# Agent Ops Coordinator
+
+## Playbook
+
+1. `pf-extractor` — agent definitions, eval scores, runs, watches, alerts, approvals.
+2. `pf-judge` — PF-AGENT (tools_allowed, eval >= 0.5, injection policy) and PF-OPS (watch breach → alert, approval gate).
+3. `pf-verifier` — run eval/publish/watch as evidence.
+4. `pf-synthesizer` — report with run/eval/alert ids.
+
+## Não faz
+- Does not publish below-gate agents; does not fire alerts without a breach.

product_forge-0.2.0/.agents/agents/data-analyst-coordinator.md

@@ -0,0 +1,18 @@
+---
+name: data-analyst-coordinator
+description: Coordinator for analytics work — sources, schemas, queries, dashboards. Use for "explore this data", "why this metric", "build a query/dashboard".
+rule_areas: [PF-SQL, PF-SRC, PF-DASH]
+skills: [pf-sql, pf-analytics, pf-dashboards]
+---
+# Data Analyst Coordinator
+
+## Playbook
+
+1. `pf-extractor` — verify sources, pull schemas, inspect stored query runs.
+2. `pf-judge` — PF-SQL (read-only, limits), PF-SRC (healthy vs unverified), PF-DASH (widget→run bindings).
+3. `pf-verifier` — execute the validated query / render the dashboard as evidence.
+4. `pf-synthesizer` — answer with run ids and row counts, never invented numbers.
+
+## Não faz
+- Never executes write/multi-statement SQL; rejected statements stay rejected.
+- Never reports unverified sources as healthy.

product_forge-0.2.0/.agents/agents/executors/pf-extractor.md

@@ -0,0 +1,16 @@
+---
+name: pf-extractor
+description: Collects facts from the Product Forge store/API — never judges. Use as first executor in any coordinator playbook.
+---
+# pf-extractor
+
+Collects `Fact`s: context packages, entity graphs, source schemas, query runs, eval scores, audit events.
+
+## Pressupõe
+- Store/API reachable; product_id known.
+
+## Entrega
+- Structured facts with ids and raw fields only — no interpretation.
+
+## Não faz
+- No judgments, no thresholds, no recommendations. Numbers without interpretation.

product_forge-0.2.0/.agents/agents/executors/pf-judge.md

@@ -0,0 +1,16 @@
+---
+name: pf-judge
+description: Applies rules/catalog/*.yaml to extracted facts — emits Findings with fact_id evidence. Use after pf-extractor.
+---
+# pf-judge
+
+Every `Finding` carries `rule_id` + non-empty `evidence[]` of fact_ids.
+
+## Pressupõe
+- Facts from pf-extractor; catalog under `rules/catalog/`.
+
+## Entrega
+- findings[] {rule_id, severity, evidence, area} — sorted deterministically.
+
+## Não faz
+- No rules not in the catalog; no findings with empty evidence; no cost estimates.

product_forge-0.2.0/.agents/agents/executors/pf-synthesizer.md

@@ -0,0 +1,14 @@
+---
+name: pf-synthesizer
+description: Composes the final answer from facts+findings+evidence — with citations and ids, advisory tone. Use last.
+---
+# pf-synthesizer
+
+## Pressupõe
+- Facts, findings, verification evidence.
+
+## Entrega
+- Answer citing entity ids, rule_ids, run ids; marks uncertain items `unresolved`.
+
+## Não faz
+- No new facts; no silent promotion of `inferred` to `observed`; no invented metrics.

product_forge-0.2.0/.agents/agents/executors/pf-verifier.md

@@ -0,0 +1,14 @@
+---
+name: pf-verifier
+description: Runs the real command to prove a claim — executes query, renders dashboard, runs eval, fires watch. Use before any "it works" statement.
+---
+# pf-verifier
+
+## Pressupõe
+- A concrete claim and the verb that proves it.
+
+## Entrega
+- Evidence block: command, run_id, status, output hash/rows.
+
+## Não faz
+- Never claims success without an execution result; never weakens criteria to pass.

product_forge-0.2.0/.agents/agents/integration-coordinator.md

@@ -0,0 +1,16 @@
+---
+name: integration-coordinator
+description: Coordinator for AWS/Azure/API/data-contract intelligence. Use for "AWS inventory", "Azure resources", "API diff", "breaking change", "data contract".
+rule_areas: [PF-CLOUD, PF-API, PF-CONTRACT]
+skills: [pf-aws, pf-azure, pf-apis, pf-data-products]
+---
+# Integration Coordinator
+
+## Playbook
+
+1. `pf-extractor` — cloud inventory states, imported API specs, contracts.
+2. `pf-judge` — PF-CLOUD (ok/unverified/error honesty), PF-API (breaking changes), PF-CONTRACT (validation errors).
+3. `pf-synthesizer` — report each item with its actual status field.
+
+## Não faz
+- NEVER fabricates cloud resources — missing credentials produce `unverified`, not inventory.

product_forge-0.2.0/.agents/agents/planning-coordinator.md

@@ -0,0 +1,17 @@
+---
+name: planning-coordinator
+description: Coordinator for discovery, prioritization, PRDs, roadmaps, decisions. Use for "prioritize", "write PRD", "roadmap", "opportunity".
+rule_areas: [PF-PLAN, PF-DEC]
+skills: [pf-prioritization, pf-requirements, pf-reporting]
+---
+# Planning Coordinator
+
+## Playbook
+
+1. `pf-extractor` — initiatives, hypotheses, decisions, knowledge citations.
+2. `pf-judge` — PF-PLAN (score inputs present, PRD citations) and PF-DEC (decisions linked to evidence).
+3. `pf-verifier` — run prioritize/PRD/roadmap and check output contract fields.
+4. `pf-synthesizer` — report with initiative/decision ids.
+
+## Não faz
+- Does not pick priorities by intuition — deterministic scoring only.

product_forge-0.2.0/.agents/agents/product-intel-coordinator.md

@@ -0,0 +1,19 @@
+---
+name: product-intel-coordinator
+description: Coordinator for product intelligence questions — decides which executor runs next. Use for "understand this product", "product context", "product health".
+rule_areas: [PF-KNOW, PF-CTX, PF-HEALTH]
+skills: [pf-knowledge, pf-product-strategy, pf-product-discovery]
+---
+# Product Intelligence Coordinator
+
+Routes work to executors. Never executes itself.
+
+## Playbook
+
+1. `pf-extractor` — pull context package, knowledge stats, entity graph.
+2. `pf-judge` — apply PF-KNOW/PF-HEALTH rules (coverage, orphan entities, stale docs).
+3. `pf-synthesizer` — compose the answer with citations.
+
+## Não faz
+- Does not query data sources or run SQL (that is data-analyst-coordinator).
+- Does not invent coverage: absent knowledge is reported as absent.

product_forge-0.2.0/.agents/skills/pf-agent-factory/SKILL.md

@@ -0,0 +1,45 @@
+---
+name: pf-agent-factory
+description: Create, evaluate and publish product agents. Use for 'create agent', 'run agent', 'agent eval', 'publish agent', 'Agent Factory'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Agent Factory
+
+## Objetivo
+Build product-scoped agents with declared tools, pass evals, gate publication.
+
+## Pré-condições
+- Product exists; knowledge imported for knowledge-backed agents
+
+## Instruções
+1. Create: `productforge agent create <pid> '<name>' --tools knowledge_query`.
+2. Run: `productforge agent run <pid> <aid> '<input>'` — deterministic + policy-checked.
+3. Eval: `productforge agent eval <pid> <aid>` → persisted score.
+4. Publish: `productforge agent publish` — requires tools_allowed AND eval score ≥ 0.5.
+
+## Ferramentas
+- `productforge agent create|run|eval|publish`
+- `GET /agents/{aid}/runs`
+- `POST /runs/{rid}/cancel`
+- `POST /workflows/run` (multi-agent)
+
+## Exemplos
+- Agent w/ knowledge_query → run 'summarize onboarding' → cited template output.
+
+## Limitações
+- Injection attempts → run status=blocked (policy engine).
+- Default runtime is deterministic template; real LLM loop (`llm_runtime.py`)
+  activates only with a configured provider subprocess — tool calls remain
+  policy-gated, tokens land in the budget ledger, no provider → honest
+  deterministic fallback (never simulated).
+
+## Evals
+- No tools_allowed → publish 400; eval 0.3 → publish rejected; score 0.7 → ok.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-agent-factory/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Agent Factory"
+  short_description: "Create, eval and publish product agents"
+  default_prompt: "Use $pf-agent-factory to create, evaluate, or publish a Product Forge agent."

product_forge-0.2.0/.agents/skills/pf-analytics/SKILL.md

@@ -0,0 +1,43 @@
+---
+name: pf-analytics
+description: Explore product data sources, schemas and analytical queries without deep SQL. Use for 'explore data', 'what tables', 'metrics', 'analytics'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Analytics
+
+## Objetivo
+Let a PM/PA answer data questions via schema explorer + compiled/NL queries.
+
+## Pré-condições
+- Data source registered and verified (`POST /data-sources/{sid}/verify` → healthy)
+
+## Instruções
+1. Verify connectivity: `POST /data-sources/{sid}/verify`.
+2. Check the metric catalog first: `GET /products/{pid}/metrics` — reuse
+   governed metrics instead of re-deriving KPIs.
+3. Browse schema: `GET /data-sources/{sid}/schema` → tables/columns/types.
+4. Ask in NL: `POST /queries/nl` — `via: semantic-catalog` (metric match) or
+   `deterministic-nl` fallback → review → execute.
+5. Persist useful runs — every execution stores a query run with provenance.
+
+## Ferramentas
+- `productforge source list|add|verify|schema`
+- `POST /queries/nl|compile|execute`
+
+## Exemplos
+- 'revenue by month' → NL scaffold SQL → execute → rows + run id.
+
+## Limitações
+- NL translation is deterministic scaffolding — always review generated SQL before execute.
+- Missing/unreachable sources report error status, never fake rows.
+
+## Evals
+- Empty table → `rows: 0` honest result; bad locator → status=error.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-analytics/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Analytics"
+  short_description: "Schema exploration and assisted queries"
+  default_prompt: "Use $pf-analytics to explore governed product data and build a safe query."

product_forge-0.2.0/.agents/skills/pf-apis/SKILL.md

@@ -0,0 +1,40 @@
+---
+name: pf-apis
+description: API intelligence — import OpenAPI/AsyncAPI, diff versions, detect breaking changes. Use for 'API spec', 'breaking change', 'API diff', 'import OpenAPI'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Apis
+
+## Objetivo
+Turn API specs into knowledge-graph entities + flag breaking changes across versions.
+
+## Pré-condições
+- OpenAPI/AsyncAPI spec file or URL
+
+## Instruções
+1. Import: `POST /products/{pid}/apis/import` {name, spec} → operations parsed to entities.
+2. List: `GET /products/{pid}/apis`.
+3. Diff: `POST /apis/{aid}/diff` {spec:new_version} → added/removed/breaking arrays.
+
+## Ferramentas
+- `POST /apis/import`
+- `GET /apis`
+- `POST /apis/{aid}/diff`
+- `GET /impact/{eid}`
+
+## Exemplos
+- v2 removes `POST /charge` → diff flags breaking removal + impact lists dependents.
+
+## Limitações
+- JSON/YAML OpenAPI 3.x + AsyncAPI 2.x; exotic $ref forms flagged, not guessed.
+
+## Evals
+- Removed endpoint → appears in `breaking`; added → `added`; graph gains operation entities.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-apis/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Apis"
+  short_description: "OpenAPI/AsyncAPI import, diff, breaking changes"
+  default_prompt: "Use $pf-apis to import an API contract and assess breaking changes."

product_forge-0.2.0/.agents/skills/pf-aws/SKILL.md

@@ -0,0 +1,39 @@
+---
+name: pf-aws
+description: AWS product intelligence — inventory, Athena/Glue/QuickSight states. Use for 'AWS resources', 'Athena', 'Glue catalog', 'QuickSight', 'AWS cost'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Aws
+
+## Objetivo
+Report AWS integration state honestly: real inventory when configured, `unverified` when not.
+
+## Pré-condições
+- AWS config/credentials present for real data
+
+## Instruções
+1. Inventory: `GET /products/{pid}/cloud/aws/inventory` — services grouped by provider.
+2. Check each service status field: `ok|unverified|error` — never present unverified as fact.
+3. On unverified: surface exactly which config is missing.
+
+## Ferramentas
+- `GET /products/{pid}/cloud/aws/inventory`
+- `GET /cloud/aws/{service}`
+
+## Exemplos
+- No creds → `{status:'unverified', missing:['AWS_PROFILE']}`.
+
+## Limitações
+- NEVER fabricate resources — absent creds = unverified. This is a hard invariant.
+- Read-only discovery; no provisioning/mutation.
+
+## Evals
+- Without creds → unverified; with bad creds → error; both recorded in audit.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-aws/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Aws"
+  short_description: "AWS product intelligence with honest states"
+  default_prompt: "Use $pf-aws to inspect AWS product intelligence with honest credential states."

product_forge-0.2.0/.agents/skills/pf-azure/SKILL.md

@@ -0,0 +1,38 @@
+---
+name: pf-azure
+description: Azure product intelligence — Fabric, Power BI, KQL workspaces. Use for 'Azure resources', 'Fabric', 'Power BI', 'KQL', 'Log Analytics'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Azure
+
+## Objetivo
+Report Azure/Fabric/PowerBI/KQL state honestly — configured vs unverified vs error.
+
+## Pré-condições
+- Azure tenant/subscription configured for real data
+
+## Instruções
+1. Inventory: `GET /products/{pid}/cloud/azure/inventory`.
+2. Per-service probes surface structured `status` + `missing` fields.
+3. Never present unverified items as discovered resources.
+
+## Ferramentas
+- `GET /products/{pid}/cloud/azure/inventory`
+- `GET /cloud/azure/{service}`
+
+## Exemplos
+- Missing tenant → `unverified` + which env vars to set.
+
+## Limitações
+- Same honesty invariant as AWS — no fabricated inventory.
+
+## Evals
+- No config → unverified with missing list; audit event recorded.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-azure/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Azure"
+  short_description: "Azure/Fabric/PowerBI intelligence, honest states"
+  default_prompt: "Use $pf-azure to inspect Azure product intelligence and configuration gaps."

product_forge-0.2.0/.agents/skills/pf-dashboards/SKILL.md

@@ -0,0 +1,42 @@
+---
+name: pf-dashboards
+description: Build and render product dashboards from real query runs. Use for 'create dashboard', 'KPI card', 'chart', 'dashboard'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Dashboards
+
+## Objetivo
+Compose KPI/chart widgets bound to executed query runs — rendered data is real data.
+
+## Pré-condições
+- ≥1 completed query run for the product
+
+## Instruções
+1. Prefer catalog metrics for KPIs — `GET /products/{pid}/metrics` gives
+   governed names + SQL; never invent a KPI formula when a catalog metric exists.
+2. Create: `POST /products/{pid}/dashboards` with widgets [{type,title,query_run_id,config}].
+3. Render: `POST /dashboards/{did}/render` → each widget resolved against its stored run.
+4. Types: `kpi`, `line`, `bar`, `table`; filters in config.
+
+## Ferramentas
+- `productforge dashboard create|list`
+- `POST /dashboards`
+- `POST /dashboards/{did}/render`
+
+## Exemplos
+- KPI 'MAU' bound to run q_abc → render returns value + refreshed_at.
+
+## Limitações
+- Widgets render from stored runs only — no live source hits at render time.
+- Missing run → widget error state, not zeros.
+
+## Evals
+- Render of 2 widgets → 2 resolved blocks; deleted run → widget shows error.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-dashboards/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Dashboards"
+  short_description: "Dashboards rendered from real query runs"
+  default_prompt: "Use $pf-dashboards to build a dashboard from real, executed query runs."

product_forge-0.2.0/.agents/skills/pf-data-products/SKILL.md

@@ -0,0 +1,39 @@
+---
+name: pf-data-products
+description: Data product contracts — define schemas, validate payloads, lineage/impact. Use for 'data contract', 'schema validation', 'lineage', 'data product'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Data Products
+
+## Objetivo
+Enforce declared data contracts and trace impact through the knowledge graph.
+
+## Pré-condições
+- Product exists
+
+## Instruções
+1. Define: `POST /products/{pid}/contracts` {name, schema{fields:[{name,type,required}]}}.
+2. Validate payload: `POST /contracts/{cid}/validate` {payload} → errors[] field-level.
+3. Impact: `GET /products/{pid}/impact/{eid}` → graph traversal of dependents.
+
+## Ferramentas
+- `POST /contracts`
+- `POST /contracts/{cid}/validate`
+- `GET /impact/{eid}`
+
+## Exemplos
+- Contract requires `user_id:string` → payload missing it → explicit error entry.
+
+## Limitações
+- Validation is structural (types/required); semantic rules are advisory.
+
+## Evals
+- Type mismatch → error naming field+expected; valid payload → `{valid:true}`
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-data-products/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Data Products"
+  short_description: "Data contracts, validation and impact"
+  default_prompt: "Use $pf-data-products to define, validate, and trace a data product contract."

product_forge-0.2.0/.agents/skills/pf-finops/SKILL.md

@@ -0,0 +1,39 @@
+---
+name: pf-finops
+description: Product FinOps — agent run budgets, cloud cost honesty. Use for 'cost', 'budget', 'FinOps', 'spend'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Finops
+
+## Objetivo
+Track agent/token budgets in the ledger and report cloud costs only when verifiable.
+
+## Pré-condições
+- Agents running (ledger auto-records); cloud config for cost probes
+
+## Instruções
+1. Budgets: `GET /products/{pid}/budgets` → ledger of per-run token/cost estimates.
+2. Cloud cost: via `GET /cloud/{aws|azure}/cost` — unverified without creds.
+3. Set routing profile so expensive models are chosen only for complex task classes.
+
+## Ferramentas
+- `GET /products/{pid}/budgets`
+- `GET /cloud/{p}/cost`
+- `POST /routing/route`
+
+## Exemplos
+- 5 agent runs → ledger shows 5 entries with model + token estimate.
+
+## Limitações
+- Estimates are declared/config-derived, not metered billing; cloud costs unverified w/o creds.
+
+## Evals
+- Each run appends a ledger entry; no creds → cost probe returns unverified.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-finops/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Finops"
+  short_description: "Agent budgets and honest cloud cost states"
+  default_prompt: "Use $pf-finops to review agent budgets and verifiable cloud cost signals."

product_forge-0.2.0/.agents/skills/pf-governance/SKILL.md

@@ -0,0 +1,39 @@
+---
+name: pf-governance
+description: Product governance — approvals, audit trail, policy gates. Use for 'approval', 'audit', 'who changed', 'governance'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Governance
+
+## Objetivo
+Keep mutating actions gated and every state change auditable.
+
+## Pré-condições
+- Product Forge API in use (audit events accumulate automatically)
+
+## Instruções
+1. Request approval: `POST /products/{pid}/approvals` {kind, subject_id, reason}.
+2. Decide: `POST /approvals/{aid}/decision` {decision:approve|reject, actor}.
+3. Inspect trail: `GET /api/v1/audit` — actor/action/resource/timestamp for every change.
+
+## Ferramentas
+- `POST /approvals`
+- `POST /approvals/{aid}/decision`
+- `GET /audit`
+
+## Exemplos
+- Publish-gated agent → approval request → approve → publish proceeds.
+
+## Limitações
+- Gates enforce within Product Forge only; external systems need their own controls.
+
+## Evals
+- Mutating call emits audit event; rejected decision blocks the gated action.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-governance/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Governance"
+  short_description: "Approvals, audit trail and policy gates"
+  default_prompt: "Use $pf-governance to gate changes with approvals and an auditable trail."

product_forge-0.2.0/.agents/skills/pf-knowledge/SKILL.md

@@ -0,0 +1,41 @@
+---
+name: pf-knowledge
+description: Knowledge management — import docs, FTS search with citations, entity graph, health. Use for 'import document', 'search knowledge', 'knowledge graph', 'context'.
+metadata:
+  version: "1.0.0"
+  forge: product-forge
+  manifest: forge/Skill/v1
+---
+# Knowledge
+
+## Objetivo
+Grow trustworthy product knowledge: chunked docs + citations + entity graph + health score.
+
+## Pré-condições
+- Product exists
+
+## Instruções
+1. Import: `productforge knowledge import <pid> <file>` or `POST /documents/import` {title,text}.
+2. Query: `productforge knowledge query <pid> '<q>'` → cited chunks (FTS5, punctuation-safe).
+3. Extract: `POST /knowledge/extract` → entities + edges into the product graph.
+4. Health: `GET /knowledge-health` → coverage/citation metrics.
+
+## Ferramentas
+- `productforge knowledge import|query|extract`
+- `GET /graph`
+- `GET /knowledge-health`
+
+## Exemplos
+- Import runbook.md → query 'rollback procedure' → top chunk + document citation.
+
+## Limitações
+- No LLM summarization w/o gateway — returns chunks, not prose answers.
+- Heuristic entity extraction — verify high-impact edges manually.
+
+## Evals
+- Imported doc retrievable by FTS with citation; health endpoint counts it.
+
+## Referências
+- API envelope: `forge/PortalApi/1` — `{correlation_id, data, error}`
+- Docs: `docs/audit/IMPLEMENTATION_GAP_MATRIX.md`, `docs/adr/ADR-001-architecture-and-persistence.md`
+- Source: `src/product_forge/`

product_forge-0.2.0/.agents/skills/pf-knowledge/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "PF Knowledge"
+  short_description: "Doc import, cited search, entity graph"
+  default_prompt: "Use $pf-knowledge to import, search, cite, and assess product knowledge."