jumpstart-mode 1.1.13 → 1.1.14

package/.jumpstart/skills/observability-design-for-ai/references/telemetry-catalog.md

@@ -0,0 +1,62 @@
+# Telemetry Catalog
+
+## Trace span fields (per LLM call)
+- correlation_id, parent_span_id, span_id
+- timestamp, duration_ms
+- caller (user_id, tenant_id, feature, app_version)
+- model (provider, model_id, model_version)
+- prompt_hash, prompt_version
+- input_tokens, output_tokens, total_cost
+- response_summary (redacted text or hash)
+- outcome (success, refused, error_type)
+- safety_flags (injection_detected, pii_filter_triggered, etc.)
+- eval_scores (when sampled)
+
+## Trace span fields (per tool call)
+- tool_name, side_effect_class
+- args (redacted), args_hash
+- result_summary (redacted)
+- upstream_status, upstream_latency_ms
+- approval_gate (none/implicit/explicit/dual), approver_id
+- error_code
+
+## Trace span fields (per retrieval)
+- retriever_id, index_version
+- query_rewrite, filters_applied
+- top_k_pre / top_k_post, reranker
+- chunk_ids returned, chunk_scores
+- ACL_filter_applied (boolean), caller_identity_used
+
+## Eval-on-production
+- sampling_rate (overall + per slice)
+- judge model and prompt
+- metrics scored: faithfulness, refusal correctness, citation correctness, task success, etc.
+- compare to release baseline; alert on drift > X%
+- write back to eval-set candidates queue
+
+## Cost attribution
+- per call cost computed from token counts × current price + tool costs.
+- aggregated by tenant, user, feature flag, prompt version, model.
+- daily budget guard with alert on burn-rate.
+- per-tenant chargeback table if multi-tenant SaaS.
+
+## Drift signals
+- retrieval recall on a held-out probe set (run on schedule).
+- calibration (predicted confidence vs. eval correctness).
+- refusal rate by slice (sudden change is a signal).
+- edit rate / override rate from HITL queues.
+- user thumbs / CSAT lagging signal.
+
+## Safety telemetry
+- prompt-injection detector hits.
+- jailbreak attempt patterns.
+- PII regex / classifier hits in inputs and outputs.
+- output classifier scores (toxicity, brand-safety).
+- abuse rate-limit triggers.
+
+Every safety hit creates a queued candidate for the eval set if it represents a new pattern.
+
+## Redaction and access
+- redact at ingest; do not rely on post-processing.
+- tier access: ops sees redacted; SREs with-privacy-training see joined; security sees raw with audit.
+- retention: per data class; default 30 days redacted, 7 days raw, longer for compliance-required.

package/.jumpstart/skills/observability-design-for-ai/templates/observability-spec.md

@@ -0,0 +1,50 @@
+# Observability Spec — `<system>`
+
+## Call types
+| Type | Span schema | Sampling |
+|---|---|---|
+
+## Trace platform
+- Backend (App Insights, OTel collector + LangSmith/Langfuse/Arize/Phoenix, Foundry):
+- Correlation strategy:
+- Schema: see `trace-schema.md`
+
+## Eval-on-production
+- Sampling rate (overall, per slice):
+- Judge model:
+- Metrics scored:
+- Drift alert thresholds:
+- Owner:
+
+## Cost attribution
+- Dimensions (tenant, user, feature, prompt version, model):
+- Aggregation cadence:
+- Budgets and burn alerts:
+
+## Drift signals
+- Retrieval probe set:
+- Calibration check:
+- Refusal/edit/override rates:
+- CSAT / thumbs feedback:
+
+## Safety telemetry
+- Injection detector:
+- PII filter:
+- Output classifiers:
+- Eval-set feedback loop:
+
+## Storage and retention
+| Stream | Store | Retention | Redaction |
+|---|---|---|---|
+
+## Dashboards and alerts
+- Pod-internal triage:
+- Sponsor trust dashboard:
+- Finance cost dashboard:
+- Alert rules and on-call:
+
+## Access control
+- Roles and what each can see:
+- Audit on access to raw content:
+
+## Risks, assumptions, next validation

package/.jumpstart/skills/observability-design-for-ai/templates/trace-schema.md

@@ -0,0 +1,40 @@
+# Trace Schema
+
+## Common fields (all spans)
+- correlation_id, parent_span_id, span_id, timestamp, duration_ms
+- tenant_id, user_id (hashed if required), session_id
+- feature, app_version, env
+
+## LLM span (`type=llm`)
+- model_provider, model_id, model_version
+- prompt_hash, prompt_version
+- input_tokens, output_tokens, cached_tokens
+- cost_usd
+- response_summary (redacted), refused (bool), error_code
+- safety_flags
+- eval_scores (when sampled)
+
+## Tool span (`type=tool`)
+- tool_name, side_effect_class
+- args_hash, args_redacted
+- upstream_status, upstream_latency_ms
+- approval_gate, approver_id
+- error_code
+- result_redacted
+
+## Retrieval span (`type=retrieval`)
+- retriever_id, index_version
+- query_rewrite, filters_applied
+- top_k_pre, top_k_post, reranker
+- chunk_ids[], chunk_scores[]
+- acl_filter_applied, caller_identity
+
+## Memory span (`type=memory`)
+- store, type (working/session/episodic/semantic/profile/cache)
+- read|write
+- record_id, scope_filter
+
+## Redaction policy
+- response_summary and args_redacted apply PII regex + classifier before persistence.
+- Raw content stored in restricted-access store with shorter retention.
+- Hashes (prompt_hash, args_hash) enable joining without exposing content.

package/.jumpstart/skills/persona-empathy-map/SKILL.md

@@ -0,0 +1,93 @@
+---
+name: persona-empathy-map
+description: 'Generate empathy maps and Jobs-To-Be-Done (JTBD) statements for Jump Start and AI Innovation Pod use-case design from interview transcripts, observation notes, or stakeholder summaries. Use this whenever the user needs structured user understanding before scoping an AI workflow, copilot, or agent — even when phrased as "build a persona", "what does the user actually need", or "give me an empathy map for the claims adjuster".'
+argument-hint: 'Provide interview transcripts, observation notes, or role descriptions plus the use case under consideration.'
+---
+
+# Persona Empathy Map
+
+## Positioning
+
+Use this skill to ground AI design in actual user reality, not assumed personas. Empathy maps + JTBD statements protect Pod and Jump Start work from solutionism by forcing the team to articulate what the user is hired to accomplish before designing the AI.
+
+## When To Use This
+
+- Pre-design phase of a copilot, agent, or workflow build.
+- Discovery synthesis after user interviews or shadowing.
+- When the team is jumping to tooling without a user lens.
+
+## Typical Need Shapes
+
+- "Build an empathy map for the claims adjuster from these 3 interview transcripts."
+- "Give me JTBD statements for the field service tech this copilot is targeting."
+- "Turn these shadowing notes into a persona we can validate."
+
+## When Not To Use This
+
+- Do not use it as a marketing persona builder. Route to `gtm-positioning-strategy` for buyer personas.
+- Do not use it to design the agent itself. Route to `agent-prd-writer` after the persona work.
+- Do not use it for stakeholder mapping. Route to `proto-persona`.
+
+## Required Inputs
+
+- Interview transcripts, observation notes, or detailed role descriptions
+- The candidate use case
+- Sample size and source (so the user knows the confidence level)
+
+## Evidence And Boundary Discipline
+
+- Cite the source quote or observation for every claim on the empathy map.
+- Distinguish observed behavior from inferred motivation.
+- If the input is thin (one interview, secondhand), say so and flag the persona as provisional.
+- JTBD statements should be specific and outcome-oriented, not feature wishlists.
+
+## Workflow
+
+1. Extract verbatim quotes about pain, gain, frustration, workaround, and measure of success.
+2. Populate the empathy map: Says / Thinks / Does / Feels.
+3. Add Pains and Gains.
+4. Derive 3–5 JTBD statements in the canonical form.
+5. Note evidence strength and unverified inferences.
+6. Surface implications for the AI design (what the agent should do, not do, and where to put HITL).
+
+## Output Structure
+
+1. Persona name and role
+2. Source summary (interviews, observations, sample size)
+3. Empathy map: Says / Thinks / Does / Feels
+4. Pains
+5. Gains
+6. JTBD statements (3–5)
+7. Evidence quality notes
+8. AI-design implications (capabilities, boundaries, HITL points)
+
+## Reference Files
+
+- Use [empathy-map-pattern.md](./references/empathy-map-pattern.md) for the canonical structure and JTBD format.
+
+## Common Pitfalls
+
+### Pitfall 1: Generic Persona
+
+Bad: "Sarah, 35, busy professional."
+
+Better: "Claims adjuster handling 40 cases/week, judged on cycle time and reopen rate, currently context-switching across 5 systems."
+
+### Pitfall 2: Inferring Feelings With No Evidence
+
+Bad: "Feels overwhelmed" with no quote.
+
+Better: tag inferred items and recommend a validation interview.
+
+### Pitfall 3: Solution-Shaped JTBD
+
+Bad: "When I open a claim, I want a copilot to summarize it."
+
+Better: "When I open a claim, I want to know what's missing or risky, so I can act in under 5 minutes without re-reading the file."
+
+## Quality Bar
+
+- Every empathy-map item is sourced or marked inferred.
+- JTBD statements are outcome-shaped, not feature-shaped.
+- Evidence strength is stated.
+- AI-design implications are explicit.

package/.jumpstart/skills/persona-empathy-map/evals/evals.json

@@ -0,0 +1,17 @@
+{
+  "skill_name": "persona-empathy-map",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "Build an empathy map and JTBD statements for the claims adjuster from these three interview transcripts. The use case is a triage copilot.",
+      "expected_output": "Sourced empathy map, 3-5 JTBD statements in canonical form, evidence quality tagged, AI-design implications surfaced.",
+      "files": []
+    },
+    {
+      "id": 2,
+      "prompt": "We have one shadowing session of a field service tech. Build a provisional persona and tell me what to validate next.",
+      "expected_output": "Persona flagged as provisional, evidence gaps named, validation plan included.",
+      "files": []
+    }
+  ]
+}

package/.jumpstart/skills/persona-empathy-map/references/empathy-map-pattern.md

@@ -0,0 +1,48 @@
+# Empathy Map Pattern
+
+## Empathy Map Quadrants
+
+Capture for each persona, with a source quote or observation per item where possible.
+
+- **Says:** verbatim things they say in interviews or on the job.
+- **Thinks:** what they believe but may not say out loud (mark inferred).
+- **Does:** observed behavior, workflows, workarounds.
+- **Feels:** emotional state and stressors (mark inferred unless directly stated).
+
+Add:
+- **Pains:** frustrations, blockers, time sinks, error sources.
+- **Gains:** outcomes they want, measures of success, what makes a good day.
+
+## JTBD Statement Format
+
+Use the canonical form:
+
+> When [situation], I want to [motivation], so I can [expected outcome].
+
+Examples:
+- "When I open a complex claim, I want to see what's missing or unusual, so I can decide whether to escalate before I burn 30 minutes."
+- "When a customer calls back about an order, I want one screen with their last 3 interactions, so I don't ask them the same questions again."
+
+JTBD anti-patterns:
+- Feature-shaped: "I want a chatbot." (Why? For what outcome?)
+- Vague: "I want to be more efficient." (Efficient at what?)
+- Tool-shaped: "I want SAP to be faster." (What is the underlying job?)
+
+## Evidence Quality Tags
+
+- **Direct quote** — verbatim from a participant.
+- **Observed** — facilitator or interviewer noted the behavior.
+- **Inferred** — reasonable read, not stated; recommend validation.
+- **Synthetic** — composite across multiple interviews; note the source count.
+
+## AI-Design Implications
+
+For each persona, surface:
+- **Capabilities the agent should have** (tied to JTBD).
+- **Boundaries / what the agent should not do** (tied to risk or trust).
+- **HITL checkpoints** (where the persona must stay in the loop).
+- **Measurement** (what good looks like for this persona — cycle time, reopen rate, accuracy).
+
+## Confidence Statement
+
+End with one sentence: how confident is this persona, based on what evidence, and what would strengthen it (more interviews, shadowing, log analysis).

package/.jumpstart/skills/persona-journey-map-companion/README.md

@@ -0,0 +1,11 @@
+# Persona Journey Map Companion
+
+Converts personas into current-state and future-state journey maps for workflow redesign and delivery planning.
+
+Typical uses:
+
+- translate persona pain points into stage-based workflow improvements
+- define where human review, approval, or escalation must remain
+- create backlog-ready improvement actions
+
+See [SKILL.md](./SKILL.md) for workflow and output contract.

package/.jumpstart/skills/persona-journey-map-companion/SKILL.md

@@ -0,0 +1,55 @@
+---
+name: persona-journey-map-companion
+description: Convert validated personas into current-state and future-state journey maps with pain points, controls, and improvement opportunities. Use this whenever the user asks to create journey maps from personas, define workflow stages per persona, or translate persona insights into implementation-ready workflow changes.
+argument-hint: Paste one or more personas and specify current-state only or current + future-state journey mapping.
+---
+
+# Persona Journey Map Companion
+
+Turn persona outputs into journey maps that support design, backlog planning, and adoption rollout.
+
+## When To Use This
+
+- Personas are finalized and workflow design must start.
+- Team needs current-state and future-state journey clarity.
+- User asks for experience maps tied to stakeholder personas.
+
+## Inputs
+
+- Persona output from user-persona-development
+- Optional process notes or workflow steps
+
+## Output Contract
+
+For each persona, produce:
+
+1. Journey stages
+2. Persona actions and decisions
+3. Pain points and delays
+4. Data or system touchpoints
+5. Risks and controls
+6. Future-state improvements
+7. Backlog candidate stories
+
+Use [journey-map-template.md](./templates/journey-map-template.md).
+
+## Workflow
+
+1. Select personas in scope (usually high and medium impact).
+2. Derive current-state stages from goals, frustrations, and workaround.
+3. Identify bottlenecks, handoffs, and approval friction.
+4. Draft future-state stage design with controls intact.
+5. Translate key changes into backlog candidates.
+
+## Quality Rules
+
+- Keep stages realistic and evidence-based.
+- Separate current-state from proposed future-state.
+- Do not collapse compliance or approval controls.
+- Keep recommendations role-specific.
+
+## Assets
+
+- Template: [journey-map-template.md](./templates/journey-map-template.md)
+- Example: [operations-journey-example.md](./examples/operations-journey-example.md)
+- Evals: [evals.json](./evals/evals.json)

package/.jumpstart/skills/persona-journey-map-companion/evals/evals.json

@@ -0,0 +1,17 @@
+{
+  "skill_name": "persona-journey-map-companion",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "Convert these two personas into current-state and future-state journey maps with backlog candidates. Keep human review controls explicit.",
+      "expected_output": "Produces stage-based current and future maps with explicit controls and actionable backlog items.",
+      "files": []
+    },
+    {
+      "id": 2,
+      "prompt": "Create a journey map from this compliance approver persona and show where approval rights must remain human-owned.",
+      "expected_output": "Retains approval boundaries and recommends improvements without removing oversight controls.",
+      "files": []
+    }
+  ]
+}

package/.jumpstart/skills/persona-journey-map-companion/examples/operations-journey-example.md

@@ -0,0 +1,29 @@
+# Journey Map: Sarah, Operations Team Lead
+
+## Persona Context
+
+- Impact level: High
+- Primary goal: identify exceptions early and keep throughput stable
+- Current workaround: shared spreadsheet with manual follow-up in Teams
+
+## Current-State Journey
+
+| Stage | Persona Action | Systems/Touches | Pain Point | Risk/Control |
+|---|---|---|---|---|
+| Intake | Reviews inbound requests | Email, CRM export | Missing fields are common | Manual validation checklist |
+| Prioritize | Sorts work by urgency | Spreadsheet | No reliable SLA indicator | Team lead judgment only |
+| Escalate | Chases dependency owners | Teams, email | Slow handoffs and unclear ownership | Escalation to manager |
+
+## Future-State Journey
+
+| Stage | Proposed Change | Enabler | Human Review/Approval | Expected Outcome |
+|---|---|---|---|---|
+| Intake | Auto-flag incomplete requests | Validation rules in API | Team lead confirms high-risk exceptions | Faster triage |
+| Prioritize | Queue with SLA risk labels | Workflow dashboard | Team lead can override ranking | Better workload balance |
+| Escalate | Structured escalation routing | Ownership map and notifications | Manager approval for escalated exceptions | Shorter cycle time |
+
+## Backlog Candidates
+
+- Add request completeness validation with severity labels.
+- Add queue filters by SLA risk and ownership.
+- Add escalation route with reason capture and audit trail.

package/.jumpstart/skills/persona-journey-map-companion/templates/journey-map-template.md

@@ -0,0 +1,29 @@
+# Journey Map: [Persona Name], [Role]
+
+## Persona Context
+
+- Impact level: [High/Medium]
+- Primary goal: [Goal]
+- Current workaround: [Workaround]
+
+## Current-State Journey
+
+| Stage | Persona Action | Systems/Touches | Pain Point | Risk/Control |
+|---|---|---|---|---|
+| [Intake] | [...] | [...] | [...] | [...] |
+| [Review] | [...] | [...] | [...] | [...] |
+| [Decision] | [...] | [...] | [...] | [...] |
+
+## Future-State Journey
+
+| Stage | Proposed Change | Enabler | Human Review/Approval | Expected Outcome |
+|---|---|---|---|---|
+| [Intake] | [...] | [...] | [...] | [...] |
+| [Review] | [...] | [...] | [...] | [...] |
+| [Decision] | [...] | [...] | [...] | [...] |
+
+## Backlog Candidates
+
+- [User story or feature candidate]
+- [User story or feature candidate]
+- [Spike or integration validation]

package/.jumpstart/skills/php-developer/README.md

@@ -0,0 +1,46 @@
+# PHP Developer Skill
+
+This skill provides a production-oriented workflow for PHP delivery in Jump Start and AI Innovation Pod engagements.
+
+## Included assets
+
+- `SKILL.md` with trigger-rich guidance and workflow
+- `references/` for framework, modernization, security, and deployment practices
+- `templates/` for bootstrap files and reusable code patterns
+- `scripts/` for lightweight assessment and modernization helpers
+- `evals/` for prompt-based quality checks
+
+## Quick start
+
+1. Run the inventory script:
+
+```bash
+python .github/skills/php-developer/scripts/inspect_php_project.py
+```
+
+2. Choose delivery mode: Laravel, Symfony, Slim/lightweight API, or legacy plain PHP.
+
+3. Apply baseline gates:
+
+```bash
+composer validate
+composer audit
+vendor/bin/phpstan analyse
+vendor/bin/phpunit
+```
+
+4. For modernization tasks, run:
+
+```bash
+python .github/skills/php-developer/scripts/legacy_risk_scan.py
+python .github/skills/php-developer/scripts/generate_refactor_backlog.py
+```
+
+## Primary scenarios
+
+- Legacy PHP stabilization and modernization
+- Laravel or Symfony service/API implementation
+- Security hardening and dependency remediation
+- Testing uplift for untested codebases
+- PHP-as-workflow-shell AI integration design
+- Production-readiness and handoff preparation