All skills
deanpeters avatar

/autonomous-investigation

@eb7a51e

The protocol behind every investigation skill. Use when AI research must proceed without you: search-plan gate, Fact/Inference/Assumption labels, confidence stacking, diffable outputs.

Use this Skill: https://skilld.dev/gh/deanpeters/product-manager-skills/autonomous-investigation

This session only. Nothing lands on disk.

template.md

≈674 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Investigation Brief Template

Copy this when launching an investigation under the autonomous-investigation protocol — either to brief an agent (paste it with the invocation) or to design a new investigation skill. It captures the contract's seven clauses as fill-in decisions, so an unattended run degrades gracefully instead of stalling or improvising.

Template

# Investigation Brief: [Target]

## The Decision
**This research supports:** [the decision — research without a decision is a hobby]
**Decision deadline:** [when — "just enough" is sized to this]

## Question Budget
**Budget:** [hard cap, usually 3]
**The questions:** [list them — anything answered inline below is already credited]
1. [Question 1]
2. [Question 2]
3. [Question 3]
**Already answered inline:** [what the invocation context already covers — don't re-ask]

## Search Plan (the gate — 3 bullets, shown before researching)
- What will be searched: [targets, topics]
- Source types: [company pages, filings, reviews, press, communities...]
- Fact/inference separation: [facts get URLs; interpretations get labeled Inference; gaps become Assumptions]

## Do-Not-Invent List (this domain's fabrication risks)
- [e.g., competitors, pricing, market share, funding rounds, customer wins, quotes]
- Rule: real, checkable URLs only — a claim without a source and date is an opinion wearing a badge

## Output Contract
**Schema:** [name the stable schema, or paste it — sections numbered, marked "do not reorder"]
**Mode:** Just Enough (Verbose only on request)
**Every key claim labeled:** Fact / Inference / Assumption — what couldn't be found goes in a gaps list, not a fourth label

## Confidence Stacking (for multi-channel findings)
- 1 channel → watch item · 2 channels → working hypothesis · 3+ → actionable · conflict → dig
- Announcements are intent until funding, hiring, procurement, or contracts corroborate

## Final Step Block
[Exactly 4 numbered next options the run will close with. On an unattended run: file the output and stop.]
1. [Option]
2. [Option]
3. [Option]
4. [Option]

## Guardrails check
- All collection is public-record OSINT: published, filed, posted, or observable
- The stage test: comfortable explaining every method at the target's user conference? If not, cut it.

Quality checks before you call it done

  • The decision is named — if it isn't, stop and get it; it defines "just enough"
  • The do-not-invent list is domain-specific, not generic
  • The schema is stable and numbered — a drifted schema silently breaks every future diff
  • The Final Step block has exactly 4 options, and the unattended behavior (file and stop) is stated

Source: SKILL.md on GitHub

1 warning2mo3 checks · Risk SAFE
  • Gen Agent Trust Hub2mo

    The skill establishes a structured protocol for autonomous research, focusing on evidence labeling (Fact, Inference, Assumption), search planning, and consistent output formatting. It includes explicit ethical guardrails and contains no executable code or malicious patterns.

  • Socket2mo

    No alerts

  • Snyk2mo

    Risk: MEDIUM · 1 issue

Signed by skilld at eb7a51e. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last month.

Activeupdated 3 months ago
type
workflow
theme
market-intelligence
estimated_time
protocol reference; investigations vary (15-45 min per run)
Other metadata
intent
Provide the canonical contract for autonomous research skills: a bounded question budget, a search-plan gate, three-level evidence labeling, do-not-invent lists, just-enough output, stable diffable schemas, and confidence stacking — so investigations are trustworthy, schedulable, and comparable run over run.
best_for
[
  "Defining consistent behavior for research skills that run as agent tasks or on schedules",
  "Keeping AI research honest: labeled evidence, real citations, no invented facts",
  "Making run N and run N+1 diffable so delta monitoring is possible"
]
scenarios
[
  "Set up a competitive scan that can re-run quarterly without me babysitting it",
  "I want research output where I can tell facts from the AI's guesses"
]

README badge

README badge for deanpeters/product-manager-skills/autonomous-investigation