القائمة
Annotates agent output with inline source citations to eliminate hallucination risk. Ensures every metric, claim, and ticket reference is traceable to a specific data source.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | cite-sources |
| version | 1.0.0 |
| description | Annotates agent output with inline source citations to eliminate hallucination risk. Ensures every metric, claim, and ticket reference is traceable to a specific data source. |
| category | quality-gates |
| trigger | Final step before delivering any agent output |
| autonomy | autonomous |
| portability | universal |
| complexity | basic |
| type | evaluation |
| inputs | [{"name":"agent_output","type":"structured-text","required":true,"description":"The full text output to annotate with source citations"}] |
| outputs | [{"name":"cited_output","type":"structured-text","description":"The agent output with inline citations added after every metric, claim, and ticket reference"}] |
| tools_optional | [] |
| model_compatibility | ["claude","gpt-4","gemini","llama-3"] |
Ensures every claim, metric, and recommendation in an agent's output is traceable to a specific data source. Eliminates hallucination risk by enforcing citation discipline.
Apply as the last quality step before delivering output, after self-check or evaluate-output. Particularly important for reports, stakeholder updates, and any output containing numbers or ticket references.
Scan the output for every number, percentage, metric, ticket reference, date, and factual claim.
For each item, attach a source citation using one of these formats:
| Claim Type | Citation Format |
|---|---|
| Ticket counts | (source: query "sprint in openSprints() AND status = Done" returned N tickets) |
| Story points | (source: story points field on PROJ-123) |
| Percentages | (source: 18 done / 24 total = 75%) |
| PR/CI data | (source: PR #456, status: merged) |
| Dates | (source: sprint end date from tracker) |
| Recommendations | (based on: 3 tickets blocked >5 days) |
Citation placement — append inline after the claim:
Velocity: 34 SP (source: sum of story points on Done tickets in Sprint 1)
Unsourceable claims — if a claim cannot be traced to provided data:
(unverified — no data source available)Citation density target: every numeric value and every factual assertion should have a citation. Narrative summaries and recommendations need at least one supporting citation.
The original agent output with inline citations added. No structural changes to the report format — citations are appended inline.
(unverified). Never silently pass unsourced claims.Generates messages suggesting a ghost-done ticket be transitioned to Done. Helpful tone, evidence-based, always asks rather than commands.
Generates contextual, humble messages designed to unblock stuck tickets. Use when a stuck ticket needs a nudge comment.
Generates a quick morning briefing with what happened, what's stuck, and what needs attention today.
Evaluates whether epics are ready for PI or quarter planning by scoring 7 readiness dimensions.
Computes team capacity for a sprint or PI from headcount, PTO, and run-rate buffer.
Estimates completion probability for remaining work using velocity distribution and Monte Carlo-style simulation.