AI-generated multi-episode works have a specific failure mode: the AI "forgets" what
happened in prior episodes. Characters lose knowledge they gained, relationships reset,
props vanish, and serialized arcs contradict themselves. This skill systematically tracks
continuity across episodes and catches every contradiction.

## Key Points

1. **Forward continuity:** Does episode N+1 honor what happened in episode N? Are
2. **Backward reference:** When episode N references a prior episode, is the reference
3. **Accumulation:** Do consequences accumulate? Injuries stack, knowledge grows,
- **Flagging intentional resets.** Some shows use time jumps, memory loss, or alternate
- **Over-tracking cosmetic details.** Focus on story-relevant continuity. A background
- **Assuming linear viewing.** Note which issues are obvious regardless of viewing order
- **Ignoring off-screen time.** Characters can learn, heal, and change between episodes.
- **Missing cumulative effects.** Track how consequences stack across episodes. AI often

Cross-Episode Continuity Checker

AI-generated multi-episode works have a specific failure mode: the AI "forgets" what happened in prior episodes. Characters lose knowledge they gained, relationships reset, props vanish, and serialized arcs contradict themselves. This skill systematically tracks continuity across episodes and catches every contradiction.

When to Use

Use when the user provides multiple episodes of a series and asks to "check continuity", "find contradictions between episodes", "track my series continuity", "verify my limited series", or when any multi-episode audit is requested. Essential for any work with 3+ episodes.

What Continuity Means Across Episodes

Continuity is the internal consistency of a story world across time. In a multi-episode work, every event in episode N constrains what can happen in episodes N+1 through the end.

Categories of Continuity

Five categories to track: Character Knowledge (who knows what and when they learned it), Relationship Status (how character pairs relate — must evolve, not reset), Physical Continuity (props, costumes, injuries, locations persist or change logically), Timeline Continuity (day/night cycles, travel times, dates add up), and World Rules (established rules stay consistent unless a rule change is a plot point).

Analysis Procedure

Step 1: Build the Episode Event Log

For each episode, create a sequential log of every significant event, tagged by category: KNOWLEDGE, RELATIONSHIP, PROP, INJURY, LOCATION, TIMELINE, RULE, ARC. Each entry records episode, scene, what changed, and who is affected.

Step 2: Cross-Reference Between Episodes

For each event logged, check all subsequent episodes for consistency:

1. Forward continuity: Does episode N+1 honor what happened in episode N? Are consequences carried forward?
2. Backward reference: When episode N references a prior episode, is the reference accurate to what actually happened in the script (not a simplified version)?
3. Accumulation: Do consequences accumulate? Injuries stack, knowledge grows, relationships evolve — they do not reset. Characters who learned something in episode 2 must still know it in episode 5.

Step 3: Track Serialized Arcs

For season-long arcs, map each beat per episode (SETUP, BUILD, ESCALATION, SETBACK, TURNING POINT, CLIMAX). Flag: beats that repeat without escalation, arc progression that reverses, missing escalation between setup and climax, and resolution contradicting established facts.

Step 4: Build the Continuity Matrix

Create a cross-episode matrix that shows the state of key elements at the end of each episode. Track every element that changes across the series. Mark any cell that contradicts a prior state as ERROR:

| Element           | EP 1 End    | EP 2 End     | EP 3 End      | EP 4 End    |
|-------------------|-------------|--------------|---------------|-------------|
| Sarah knows about M | Yes        | Yes          | Yes           | ERROR: No?  |
| Sarah's car       | Destroyed   | New (unexpl) | New           | New         |
| Sarah/Marcus rel  | Hostile     | Reconciling  | Allied        | ERROR: Reset|
| Elena's location  | New York    | New York     | ERROR: London | London      |

Any cell marked ERROR represents a continuity break that must be resolved. The matrix makes patterns visible — if one character or element has multiple ERRORs, it likely indicates a systemic AI context failure for that thread.

Step 5: Identify AI-Specific Continuity Failures

Three predictable AI failure patterns: Context window amnesia (knowledge resets, relationships revert, world details change, personalities shift between episodes), Template bleeding (same scene structure repeated, identical reactions, recycled dialogue), and Summary drift (broad strokes correct but specific details wrong, events referenced slightly differently than written, timeline inconsistencies).

Continuity Severity Levels

Severity	Definition	Example
CRITICAL	Breaks the plot; audience will notice immediately	Dead character appears alive with no explanation
MAJOR	Significant inconsistency; attentive viewers catch it	Character uses info they should not have
MINOR	Small detail inconsistency; only careful readers notice	Prop in wrong location; minor timeline gap
COSMETIC	Extremely minor; fixable in production	Costume detail inconsistency; background element

Output Format

Produce: Episode Event Log (condensed, with episode/scene/category/detail), Continuity Matrix (cross-episode state table with ERROR markers), Issues List (each with severity, what was established, what contradicts it, category, likely AI cause, and fix), Serialized Arc Tracker (beat progression with gaps/reversals), and AI Pattern Analysis (counts of context amnesia, template bleeding, and summary drift instances).

Anti-Patterns

• Flagging intentional resets. Some shows use time jumps, memory loss, or alternate timelines that intentionally reset continuity. Confirm with the user before flagging.
• Over-tracking cosmetic details. Focus on story-relevant continuity. A background extra's shirt color is not worth tracking unless plot-relevant.
• Assuming linear viewing. Note which issues are obvious regardless of viewing order vs. which only matter in sequential viewing.
• Ignoring off-screen time. Characters can learn, heal, and change between episodes. The question is whether off-screen change is plausible given the time gap.
• Missing cumulative effects. Track how consequences stack across episodes. AI often treats each episode as a fresh start, ignoring accumulated damage, knowledge, and relationship changes from prior episodes.