Blindspot

blindspot — hunt the unknown-unknowns

The most valuable thing this skill produces is the gap the user could not have asked about. Generic checklists fail twice: they list things he already has (wasting his trust) and miss the specific silent killer in HIS system. Everything below exists to prevent those two failures.

The method

1. Ground in what actually exists — before any opinion

Read the real system first: code, configs, docs, runbooks, git history, live state. Build a one-paragraph inventory of what's already covered. A gap claim that turns out to already exist destroys the report's credibility; a recommendation list that never says "you already have X covered" reads as unread boilerplate. Budget real effort here — in a repo, that means opening files, not skimming the README. If ultracode/Workflow is available and the system is large, fan out readers per subsystem and adversarially verify candidate gaps before reporting (a "gap" a verifier can refute with a file reference was never a gap).

2. Walk the expert lenses

The core mechanic: examine the system as a series of domain veterans would, each asking one question — "what does every mature version of this thing have that this one doesn't?" Pick the 5–8 lenses that fit the domain (add your own; these are seeds, not a form to fill):

For each lens, name the SPECIFIC missing thing in THIS system with evidence (file, behavior, absence), never the abstract practice. "You have no staged rollout" is noise; "every user's launcher auto-pulls main, so a bad release hits the whole fleet within hours and only a manual git revert saves them — there is no remote latest-good manifest" is a finding.

3. Rank by consequence, not interestingness

Order findings by (user impact × irreversibility × probability), and label each with its class:

• Silent killer — breaks users with no triggering event, nothing to roll back
• One-way door — cheap to fix today, a permanent regret in a year
• Single point of failure — one machine, one token, one person
• Erosion — degrades trust gradually (docs drift, flaky installs, stale mirrors)

Drop anything that is merely "best practice" without a concrete failure story in this system. If it can't hurt the user's users, it doesn't make the list.

4. Deliver the contract

Lead with the shortlist — the reader sees one screen first. Structure:

1. Build-now shortlist (≤5): each one sentence — the gap, the concrete failure it prevents, rough cost now vs. cost later. Flag any item that is a new feature if a no-new-features rule is in force (check memory — such items may need to route through an experiment lane instead of mainline).
2. Ranked findings with class labels and evidence.
3. Already covered — 3–6 things checked and found solid, with what covers them. This section is what makes the rest believable.
4. One-way doors closing soon — decisions to make deliberately this month even if the answer is "keep as is".

Failure modes to avoid

• Proposing what already exists (you didn't read enough — go back to step 1).
• The consultant dump: twenty framework-shaped bullets with no file paths.
• Overfitting to the last domain you analyzed — re-derive the lens set fresh for each system (a release train, a sales deck, and a family estate plan share almost no lenses).
• Ranking by intellectual interest instead of user pain.
• Hedging the shortlist. The user is explicitly delegating judgment — commit to an ordering and say why #1 is #1.

Run this — do not improvise

This capability's deterministic implementation is a RAPP single-file agent, linked beside this file as blindspot_agent.py and embedded as the fenced Python below (sha256 c611671d1d5a0616…; a byte-exact copy is also vaulted in the capsule comment at the end of this file). On a host with sandbox execution, run the linked file directly — if it is missing, write the fence contents verbatim to blindspot_agent.py first:

python3 blindspot_agent.py '{"key": "value"}'      # arguments as one JSON object
echo '{"key": "value"}' | python3 blindspot_agent.py   # or on stdin
python3 blindspot_agent.py --tool                      # emit the JSON tool contract

Treat stdout as a tool result. If it reports missing or unresolved inputs, stop and collect them. If it returns steps, execute those steps in order exactly as returned; if it returns instructions, follow them with the supplied inputs. Otherwise use the result verbatim. Do not invent behavior beyond that output. On a host without code execution, treat the Parameters schema and the code below as the exact specification and never paraphrase a step. Never edit inside the generated markers; a converter-equipped host can instead restore the original file checksum-verified with the installed rapp-agent-converter/scripts/toast.py convert SKILL.md --to agent.

"""Blindspot -- Systematically surface what the user doesn't know they don't know — the gaps, risks, and missing practices in a system, plan, or product only visible from cross-domain expertise. Use on "what am I missing", "blindspot this", "what would an expert see here", "stress test my plan", "sanity check this", or before committing to something hard to reverse (a launch, release process, schema, pricing, architecture). Deliverable: a ranked, evidenced gap list plus a build-now shortlist, not a checklist.

Generated by the rapp skill from blindspot. The RCI capsule at the bottom of this file carries the full original; `toast.py convert` restores it byte-exact."""

import json
import re
import sys

try:
    from agents.basic_agent import BasicAgent
except ImportError:  # running OUTSIDE a brainstem -- stay executable anyway.
    class BasicAgent:  # noqa: D101 - minimal stand-in, same contract
        def __init__(self, name=None, metadata=None):
            if name:
                self.name = name
            if metadata:
                self.metadata = metadata

        def perform(self, **kwargs):
            return "Not implemented."

        def system_context(self):
            return None

        def to_tool(self):
            return {"type": "function", "function": {
                "name": self.name,
                "description": self.metadata.get("description", ""),
                "parameters": self.metadata.get("parameters", {})}}

# The procedural layer, verbatim from the source capability.
INSTRUCTIONS = '# blindspot — hunt the unknown-unknowns\n\nThe most valuable thing this skill produces is the gap the user could not have\nasked about. Generic checklists fail twice: they list things he already\nhas (wasting his trust) and miss the specific silent killer in HIS system.\nEverything below exists to prevent those two failures.\n\n## The method\n\n### 1. Ground in what actually exists — before any opinion\n\nRead the real system first: code, configs, docs, runbooks, git history, live\nstate. Build a one-paragraph inventory of what's already covered. A gap claim\nthat turns out to already exist destroys the report's credibility; a\nrecommendation list that never says "you already have X covered" reads as\nunread boilerplate. Budget real effort here — in a repo, that means opening\nfiles, not skimming the README. If ultracode/Workflow is available and the\nsystem is large, fan out readers per subsystem and adversarially verify\ncandidate gaps before reporting (a "gap" a verifier can refute with a file\nreference was never a gap).\n\n### 2. Walk the expert lenses\n\nThe core mechanic: examine the system as a series of domain veterans would,\neach asking one question — *"what does every mature version of this thing\nhave that this one doesn't?"* Pick the 5–8 lenses that fit the domain (add\nyour own; these are seeds, not a form to fill):\n\n| Lens | The veteran's reflex question |\n|---|---|\n| Release engineering | What happens when a bad version ships — can it self-heal without users acting? |\n| Silent killers | What breaks every user at once with NO release event (unpinned deps, expiring certs/tokens, upstream API sunsets)? |\n| Data durability | What user data exists, what versions its schema, what migrates it, what backs it up, what restores it? |\n| Security / supply chain | What runs code or installs packages on trust? What's signed, what's merely fetched? |\n| Ecosystem & compatibility | What contract do third parties (plugin/agent/API authors) rely on, and what enforces it across versions? |\n| Incident response | When it breaks at 2am, where do users look, and what's the rehearsed rollback? |\n| Bus factor | What knowledge or access exists only in one person's head or machine? |\n| Observability | When a user says "it's broken", what evidence can they hand over — and what does it leak? |\n| Growth / adoption | What makes the second user's experience worse than the first's (scaling cliffs, support load, docs drift)? |\n| One-way doors | Which of today's casual choices (names, formats, schemas, URLs) become expensive-forever contracts? |\n\nFor each lens, name the SPECIFIC missing thing in THIS system with evidence\n(file, behavior, absence), never the abstract practice. "You have no staged\nrollout" is noise; "every user's launcher auto-pulls main, so a bad release\nhits the whole fleet within hours and only a manual git revert saves them —\nthere is no remote latest-good manifest" is a finding.\n\n### 3. Rank by consequence, not interestingness\n\nOrder findings by (user impact × irreversibility × probability), and label\neach with its class:\n\n- **Silent killer** — breaks users with no triggering event, nothing to roll back\n- **One-way door** — cheap to fix today, a permanent regret in a year\n- **Single point of failure** — one machine, one token, one person\n- **Erosion** — degrades trust gradually (docs drift, flaky installs, stale mirrors)\n\nDrop anything that is merely "best practice" without a concrete failure story\nin this system. If it can't hurt the user's users, it doesn't make the list.\n\n### 4. Deliver the contract\n\nLead with the shortlist — the reader sees one screen first. Structure:\n\n1. **Build-now shortlist** (≤5): each one sentence — the gap, the concrete\n   failure it prevents, rough cost now vs. cost later. Flag any item that is\n   a new feature if a no-new-features rule is in force (check memory — such\n   items may need to route through an experiment lane instead of mainline).\n2. **Ranked findings** with class labels and evidence.\n3. **Already covered** — 3–6 things checked and found solid, with what\n   covers them. This section is what makes the rest believable.\n4. **One-way doors closing soon** — decisions to make deliberately this\n   month even if the answer is "keep as is".\n\n## Failure modes to avoid\n\n- Proposing what already exists (you didn't read enough — go back to step 1).\n- The consultant dump: twenty framework-shaped bullets with no file paths.\n- Overfitting to the last domain you analyzed — re-derive the lens set fresh\n  for each system (a release train, a sales deck, and a family estate plan\n  share almost no lenses).\n- Ranking by intellectual interest instead of user pain.\n- Hedging the shortlist. The user is explicitly delegating judgment — commit to\n  an ordering and say why #1 is #1.'

# Ordered commands lifted verbatim from the capability's own documentation.
STEPS = []


class BlindspotAgent(BasicAgent):
    def __init__(self):
        self.name = 'Blindspot'
        self.metadata = {
          "name": "Blindspot",
          "description": "Systematically surface what the user doesn't know they don't know \u2014 the gaps, risks, and missing practices in a system, plan, or product only visible from cross-domain expertise. Use on \"what am I missing\", \"blindspot this\", \"what would an expert see here\", \"stress test my plan\", \"sanity check this\", or before committing to something hard to reverse (a launch, release process, schema, pricing, architecture). Deliverable: a ranked, evidenced gap list plus a build-now shortlist, not a checklist.",
          "parameters": {
            "type": "object",
            "properties": {},
            "required": []
          }
        }
        super().__init__(name=self.name, metadata=self.metadata)

    def perform(self, **kwargs):  # toaster:generated-perform
        return json.dumps({"status": "ok", "instructions": INSTRUCTIONS,
                           "inputs": kwargs,
                           "note": "Prose-only capability: follow INSTRUCTIONS "
                                   "with the given inputs."}, indent=2)

if __name__ == "__main__":
    #     echo '{"arg": "value"}' | python3 blindspot_agent.py
    #     python3 blindspot_agent.py '{"arg": "value"}'
    #     python3 blindspot_agent.py --tool          # emit the JSON tool contract
    _a = sys.argv[1:]
    if _a and _a[0] == "--tool":
        print(json.dumps(BlindspotAgent().to_tool(), indent=2))
    else:
        _raw = _a[0] if _a else (sys.stdin.read().strip() or "{}")
        print(BlindspotAgent().perform(**json.loads(_raw)))

# rci-capsule:v1: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