Competitive Intelligence Digest Agent AgentKit released the Competitive Intelligence Digest Agent, an open-source AI agent that compiles competitor digests from public sources while labeling items as confirmed facts or unverified rumors. The agent includes an AgentAz governance specification for security review, defining trust levels, tool boundaries, and human handoff triggers. The release aims to provide a safe, auditable framework for competitive research without fabricating data. Overview Compiles a competitor digest — product, pricing, positioning, news — with every item cited and dated. Labels each item as confirmed fact or unverified rumor, so nothing reads as more certain than it is. Draws only from public, ethical sources — no leaks, no non-public data, no fabricated moves. Defensive: flags low-confidence items, notes recency, and marks assumptions rather than guessing. AgentAz™ specification A lightweight, design-time governance spec for security review. It documents what this agent is authorized to do — and why — and pairs with whatever policy engine you already run. It does not enforce anything at runtime. Machine-readable contract agentaz.json , validated against the open AgentAz™ JSON Schema — bundled for offline use and published at a permanent URL: { "$schema": "./agentaz.schema.json", "version": "2.0.0", "last reviewed": "2026-06-24", "agent id": "competitive-intel-agent", "trust level": "A1", "dna pattern": "Research", "worst case action": "Includes a stale fact in a digest for human review. Never publishes or acts.", "authority boundary": "Gathers and cites competitive intel; never fabricates; publish/send tools absent.", "tags": "research", "competitive-intel", "read-only", "cited" , "tool boundary": { "allowed tools": "search sources", "gather signals", "cite source", "flag stale" , "execution tools absent": true, "read only": true }, "output boundary": { "format": "structured json", "never emits": "publish", "send" , "never fabricates": true }, "cost boundary": { "max usd per trace loop": 0.3, "alert threshold usd": 0.2 }, "loop boundary": { "max reasoning turns": 10 }, "human handoff": { "triggers": "stale data", "conflicting sources" , "destination": "analyst" }, "audit": { "append only": true, "logs": "sources", "citations" } } New to this? Read the AgentAz specification guide /agentaz-specifications — Trust Levels, DNA patterns, and how it complements your runtime. AgentAz™ is open source under Apache-2.0 https://www.apache.org/licenses/LICENSE-2.0 — schema frozen v1.0.0 and source on GitHub https://github.com/agent-kits/agentaz . Governance matrix A scannable summary of this blueprint's governance coverage, derived from its AgentAz™ specification. It documents the boundaries that already ship — not new functionality. | Agent goal | Bounded by the authority spec above | |---|---| | Trust Level | A1 — Research | | Tool access | Least privilege — execution tools absent read-only | | Context handling | Grounded in provided inputs; cites or flags rather than guessing | | Memory strategy | Task-scoped; no persistent cross-session memory | | Human approval | Required on stale data, conflicting sources → analyst | | Audit trail | Append-only log sources, citations | | Cost & loop bounds | ≤ $0.3 per loop · ≤ 10 reasoning turns | | Recovery / escalation | Escalates to analyst | Agent component mapping A framework-neutral view of how this blueprint maps to standard agent-architecture components the vocabulary common to ADK-style frameworks . It describes structure for clarity — not an official integration or certified compatibility. | Agent | Primary reasoner — Research authority A1 | |---|---| | Tools | search sources, gather signals, cite source, flag stale — execution tools absent read-only | | Memory | Task-scoped working context; no persistent cross-session memory | | Guardrails | Worst-case classified A1 ; no execution tools; ≤ $0.3/loop · ≤ 10 turns | | Evaluator | Confidence and authority-boundary checks; low-confidence or out-of-bounds results are flagged, not actioned | | Handoff | Escalates to analyst on stale data, conflicting sources | Failure modes Specific ways this blueprint can fail, and how it is designed to detect, contain, and recover from each — the boundaries that make it safe to run, stated plainly. Includes a stale or wrong fact in the digest that a reader acts on. - Detection - Every claim is cited and stale or unverifiable items are flagged. - Mitigation - It gathers and cites public information only; it never publishes or acts. - Recovery - The reader verifies against the citation and discards the bad item. Fabricates a competitor claim or metric. - Detection - Uncited claims are withheld rather than asserted. - Mitigation - It never invents a competitor claim or figure. - Recovery - The analyst confirms against the source. Presents a single source as settled fact despite conflicts. - Detection - Conflicting sources are flagged. - Mitigation - It surfaces the conflict rather than resolving it silently. - Recovery - The analyst reconciles the sources. Evaluation Factual accuracy with citations is primary — a stale or fabricated competitor fact that a reader acts on is the failure. | Fact accuracy | Share of digest claims that are correct and current versus verified sources. | |---|---| | Citation rate | Share of claims that cite a source, with uncited claims withheld. | | Fabrication rate | Frequency of invented competitor claims or metrics — should be near zero. | | Conflict surfacing | Share of conflicting sources flagged rather than silently resolved. | | Latency | Time to compile a digest. | Recommended approach. Use a set of digests with verified reference facts; measure accuracy and citation rate and audit for fabrication. Include topics with conflicting sources to confirm conflicts are surfaced, not smoothed over. When to use Use it when - You track competitors and want a regular, cited digest instead of scattered notes. - You have public sources sites, news, filings, changelogs the agent can monitor and cite. - You want confirmed facts clearly separated from rumor and speculation. - You want an ethical, defensible intel process with sources attached. Avoid it when - You want it to obtain or use non-public, leaked, or illicitly sourced information — it won't. - You expect it to predict a competitor's roadmap as fact it labels inference as inference . - You have no sources for it to ground claims in. - You need legal/strategic decisions made for you it informs; humans decide . System prompt You are a Competitive Intelligence Digest Agent. You compile a digest of competitor activity product, pricing, positioning, news from PUBLIC sources, for an internal team. You are judged on a useful, accurate, source-cited, ethically-sourced digest and on never fabricating competitor moves or presenting rumor as fact. == CORE PRINCIPLES == 1. Cite or don't claim. Every item must cite a public source you actually consulted, with a date. No source = it doesn't go in as fact. 2. Fact vs. rumor vs. inference. Clearly label each item: confirmed sourced fact , unconfirmed rumor/unverified report , or inference your hypothesis . Never blur these. A rumor stays a rumor. 3. Ethical sourcing only. Use public information websites, news, filings, changelogs, public social posts . Never use, request, or speculate from non-public, leaked, confidential, or illicitly obtained information. == HARD RULES NON-NEGOTIABLE == - NO FABRICATION: Never invent a competitor launch, price, hire, or statement. If you can't source it, it's not in the digest or it's clearly an inference . - NO NON-PUBLIC DATA: Do not use or seek trade secrets, leaked docs, private communications, or anything obtained unethically. Public sources only. - LABEL CONFIDENCE: Mark each item confirmed / unconfirmed / inference, with a source and date. Flag stale data. - NO DECISIONS: You inform strategy; you don't make competitive or legal decisions. Avoid legal-risk framing e.g. don't advise on anti-competitive actions . - RECENCY: Note how recent each item is and flag when something may be outdated. == METHOD == - For each competitor, gather from public sources. Classify each finding confirmed/unconfirmed/inference , cite and date it, and assess significance. Surface what changed since the last digest. == OUTPUT FORMAT return ONE JSON object == { "period": "