gooseworks-ai/goose-skills / skills/video/render-street-interview
render-street-interview
VerifiedWrite a street-interview ad from a complete inspected commercial interaction and current brand facts. Render product guessing, or prepare mic-only, prepared-sample or visible-task conversation script/prompt previews. Premise, actions and ad connection vary; capture rules stay fixed. Conversation media delivery is unverified.
scripts/conversation.py
View on GitHub"""Street conversation script/prompt previews; rendered delivery is unverified."""
import re
MODE = "conversation"
SHOT_KINDS = ("question", "answer", "followup", "reaction", "action")
INTERACTION_TYPES = ("mic-only", "product-sample", "concept-challenge")
CLAUSES = {
"execution mode: conversation": "record the supported execution explicitly",
"only the interviewer holds the microphone": "keep one mic and two distinct voices",
"they look at the interviewer while they speak": "participants respond to a person",
"nobody says any word that is not written below": "generate only approved spoken words",
"one continuous street ambience": "keep audio in the same space",
"every shot is filmed on one corner": "bind location across cuts",
"no on-screen text, no subtitles": "draw brand graphics after generation",
"never blurred and never bokeh": "keep the street in focus; blurred backgrounds are rejected",
}
INTERACTION_CLAUSES = {
"mic-only": {
"no product, phone or screen is held or demonstrated": "the default needs no prop",
},
"product-sample": {
"the participant handles only the prepared sample": "separate sample and microphone",
"samples are already prepared before the first shot": "do not imply package-opening support",
"no phone, screen or ui is held or demonstrated": "no device or UI support is verified",
},
"concept-challenge": {
"perform only the visible task described below": "keep the configured challenge visible",
"no phone, screen or ui is held or demonstrated": "no device or UI support is verified",
},
}
PER_SHOT = ("same street corner", "same participant", "microphone")
PRODUCT_FLAGS = ("guards", "can", "can_size", "can_sealed", "upright", "mic_ref")
_DEVICES = r"phones?|smartphones?|screens?|ui|interfaces?|laptops?|tablets?"
_SAMPLES = r"samples?|drinks?|bottles?|cups?|sip(?:s|ping)?|tast(?:e|es|ing)|(?:a|the|sealed|open) can|(?:holds?|holding|shows?|showing|handles?|handling|hands?|passing|displays?) (?:a |the |their )?products?"
def interaction_type(cfg):
"""Old configs remain mic-only. Interaction data describes a preview, never bindings."""
return cfg.get("interaction", {}).get("type", "mic-only")
def _has_visible_term(text, terms):
# Negative instructions such as "no phone or screen" may restate the scaffold's guard.
# Check visual descriptions, not dialogue: discussing a device is not demonstrating one.
text = re.sub(r"\b(?:no|without)\s+(?:(?:a|an|any)\s+)?(?:" + terms + r")"
r"(?:\s*[,/]?\s*(?:or|and)?\s*(?:" + terms + r"))*", "", text.lower())
return bool(re.search(r"\b(?:" + terms + r")\b", text))
def validate(cfg):
if not isinstance(cfg, dict):
raise SystemExit("conversation config must be an object")
for key in ("brand", "slug", "location", "question", "shots", "generation", "participant"):
if not cfg.get(key):
raise SystemExit(f"conversation config needs {key}")
for key in ("brand", "slug", "question", "participant"):
if not isinstance(cfg[key], str) or not cfg[key].strip():
raise SystemExit(f"conversation {key} must be nonempty text")
loc, gen = cfg["location"], cfg["generation"]
if not isinstance(loc, dict) or not isinstance(gen, dict):
raise SystemExit("conversation location and generation must be objects")
if not all(isinstance(loc.get(k), str) and loc[k].strip() for k in ("description", "landmarks")):
raise SystemExit("conversation location needs description and landmarks")
if (not isinstance(gen.get("seed"), int) or isinstance(gen.get("seed"), bool)
or not isinstance(gen.get("duration"), (int, float))
or not 6 <= gen["duration"] <= 15):
raise SystemExit("conversation generation needs a seed and duration from 6 to 15 seconds")
if cfg.get("product") or cfg.get("episode_role") or gen.get("answers_only"):
raise SystemExit("conversation is a standalone preview, without product reference or episode roles")
if any(gen.get(k+"_grammar") for k in PRODUCT_FLAGS):
raise SystemExit("product/reference grammar is incompatible with conversation previews")
interaction = cfg.get("interaction")
if "interaction" in cfg:
if not isinstance(interaction, dict) or interaction.get("type") not in INTERACTION_TYPES:
raise SystemExit("conversation interaction needs a supported type")
for key in ("visible_setup", "participant_reason"):
if not isinstance(interaction.get(key), str) or not interaction[key].strip():
raise SystemExit(f"conversation interaction needs nonempty {key}")
if "props" in interaction and (not isinstance(interaction["props"], str)
or not interaction["props"].strip()):
raise SystemExit("conversation interaction.props must be a nonempty description")
mode = interaction_type(cfg)
shots = cfg["shots"]
if not isinstance(shots, list) or not 3 <= len(shots) <= 8:
raise SystemExit("conversation needs 3 to 8 ordered shots")
if not all(isinstance(shot, dict) for shot in shots):
raise SystemExit("conversation shots must be objects")
speakers = set()
for shot in shots:
kind, speaker = shot.get("kind"), shot.get("speaker")
if kind not in SHOT_KINDS or speaker not in ("interviewer", "participant"):
raise SystemExit("conversation shot kind or speaker is invalid")
if kind in ("question", "followup") and speaker != "interviewer":
raise SystemExit("question/followup belongs to the interviewer")
if kind == "answer" and speaker != "participant":
raise SystemExit("answer belongs to the participant")
if not isinstance(shot.get("line", ""), str):
raise SystemExit("conversation line must be text")
if kind in ("reaction", "action") and shot.get("line", "").strip():
raise SystemExit("reaction/action shots are silent; put speech in an answer or followup")
if kind not in ("reaction", "action") and not shot.get("line", "").strip():
raise SystemExit("spoken shots need a line")
for key in ("action", "visual", "manner"):
if key in shot and not isinstance(shot[key], str):
raise SystemExit(f"conversation shot {key} must be text")
if kind == "action" and not any(shot.get(k, "").strip() for k in ("action", "visual")):
raise SystemExit("silent action shots need an explicit action or visual description")
if shot.get("line", "").strip():
speakers.add(speaker)
questions = [i for i, shot in enumerate(shots) if shot["kind"] == "question"]
if not questions or shots[questions[0]]["line"] != cfg["question"]:
raise SystemExit("question metadata must exactly mirror the first interviewer question")
if any(shot["kind"] not in ("answer", "action", "reaction") for shot in shots[:questions[0]]):
raise SystemExit("a cold open must be a participant answer or silent action/reaction")
question_words = " ".join(cfg["question"].split()).casefold()
if sum(" ".join(shot.get("line", "").split()).casefold().count(question_words)
for shot in shots) != 1:
raise SystemExit("the first interviewer question must be performed exactly once")
if speakers != {"interviewer", "participant"}:
raise SystemExit("both people must speak")
visuals = [interaction.get("visible_setup", ""), interaction.get("props", "")] if interaction else []
visuals += [shot.get(key, "") for shot in shots for key in ("action", "visual", "manner")]
if any(_has_visible_term(text, _DEVICES) for text in visuals):
raise SystemExit("conversation previews do not support visible phones, screens or UI")
if mode == "mic-only" and any(_has_visible_term(text, _SAMPLES) for text in visuals):
raise SystemExit("mic-only conversation cannot show product or sample handling")
if sum(len(s.get("line", "").split()) for s in shots) > gen["duration"] * 2.5:
raise SystemExit("conversation exceeds provisional 2.5 words/second; leave time for replies")
def build_prompt(cfg):
validate(cfg)
loc = cfg["location"]
shots = cfg["shots"]
mode = interaction_type(cfg)
interaction = cfg.get("interaction")
prop_rules = {
"mic-only": "No product, phone or screen is held or demonstrated. The participant's "
"hands rest normally unless an explicit non-prop action is supplied. ",
"product-sample": "The participant handles only the prepared sample; the interviewer "
"holds only the microphone. Samples are already prepared before the "
"first shot; no package opening is performed. No phone, screen or UI "
"is held or demonstrated. Sample appearance and handling are text "
"directions for this preview; no product reference image is bound. ",
"concept-challenge": "Perform only the visible task described below, using its described "
"non-UI props. No phone, screen or UI is held or demonstrated. ",
}
parts = [
f"EXECUTION MODE: conversation. INTERACTION TYPE: {mode}. Raw vertical street-interview footage, a stable medium "
f"shot at eye level, {loc.get('light', 'ordinary daylight')}. One adult participant and an interviewer just "
"outside frame. ONLY THE INTERVIEWER HOLDS THE MICROPHONE: one plain black handheld "
"reporter's mic, visible below the participant's chin, held from the lower right. "
+ prop_rules[mode] + "The same participant stays on screen during both voices. "
"THEY LOOK AT THE INTERVIEWER WHILE THEY SPEAK. The interviewer asks and listens; "
"the participant answers. Two distinct nearby voices recorded by the same mic. "
"Relaxed real-time pace with room between turns; perform the supplied words without "
"adding laughs or verbal filler. NOBODY SAYS ANY WORD THAT IS NOT WRITTEN BELOW. "
"When the other person speaks, lips remain closed. People and traffic move in the "
"background. The street behind them stays in focus and readable as a real place, never "
"blurred and never bokeh. No legible clothing logos or signage. No on-screen text, no subtitles. "
f"EVERY SHOT IS FILMED on ONE corner: {loc['description']}. "
f"{loc['landmarks']} stay behind the participant. "
f"Participant: {cfg['participant']}. THE {len(shots)} SHOTS: "
]
if interaction:
parts.insert(0, f"Visible setup: {interaction['visible_setup']} "
f"Reason the participant joins: {interaction['participant_reason']} "
+ (f"Configured props: {interaction['props']} " if interaction.get("props") else ""))
if shots[0]["kind"] != "question":
parts.insert(0, "Edited cold open: perform the supplied shot order. The first answer or "
"silent action/reaction precedes the question in the edit. Do not move "
"the question to the opening or repeat it. ")
for i, shot in enumerate(shots, 1):
line = shot.get("line", "")
if not line.strip():
line = ""
speech = (f"The {shot['speaker']} says exactly {line!r}. " if line else
f"Silent {shot['kind']} by the {shot['speaker']}; both voices are silent. ")
description = " ".join(shot.get(k, "") for k in ("visual", "action", "manner") if shot.get(k))
parts.append(f"{i}. Same street corner: {loc['description']}; same participant, "
f"same microphone below the chin. {speech}{description} ")
parts.append("Sound: close voices and one continuous street ambience, traffic and "
"footsteps, no music or score. Branded graphics and disclosure are added locally.")
return "".join(parts)
def lint(prompt, split_shots):
low = prompt.lower()
errors = [f'conversation prompt is missing "{k}" -- {v}' for k, v in CLAUSES.items() if k not in low]
kinds = re.findall(r"interaction type: ([a-z-]+)\.", low)
if len(kinds) != 1 or kinds[0] not in INTERACTION_TYPES:
errors.append("conversation prompt must name one supported interaction type")
else:
errors += [f'conversation prompt is missing "{k}" -- {v}'
for k, v in INTERACTION_CLAUSES[kinds[0]].items() if k not in low]
shots = split_shots(prompt)
if not shots:
errors.append("conversation has no numbered shot list")
for i, shot in enumerate(shots, 1):
for needle in PER_SHOT:
if needle not in shot.lower():
errors.append(f"conversation shot {i} misses {needle}")
if "@image" in low:
errors.append("conversation preview must not refer to an unbound image")
return errors
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |