# AGENTS.md — Build guide for Ghostfeed

## Project scope
Prepare a short user-owned clip, generate an optional identity-consistent first frame from an authorized likeness and require approval before requesting a paid motion render. Burn reviewed hook text into a separate export.

Catalogue verdict: kinda. No model here was trained by Ghostfeed · every step is somebody else's hosted inference, spread across four vendors. A Gemini image-edit call swaps your avatar into frame 0 of an existing clip, an approval gate makes you accept that frame before anything expensive runs, and then the video step splits: the default clone family sends the approved frame plus the original clip to Kling motion-control, so the render inherits the source video's motion, while the prompt family animates the still alone on PixVerse, Seedance, Grok or Kling. Rebuild the loop for yourself with a fal.ai key, a Gemini key and ffmpeg over a long weekend and it works. What does not fall out of that weekend is the rest of it · a template library scraped and scene-cut into 9:16 clips, a director that writes slideshow copy and casts every background itself, a timeline editor with its own render workers, 44 agent-API tools, and a TikTok app that took platform review to get. Personal version, yes. The thing you would actually run daily, no.
Use the implementation prompt below to define the deliverable. Complete each phase's acceptance checks before extending the scope.

## Working agreement
- Inspect the repository and its existing instructions before choosing paths, dependencies or commands. Keep one coherent stack and explain changes to the proposed architecture.
- Plan a vertical slice that accepts a real input and produces the useful output described below. Persist only the state the prompt calls for; respect memory-only and upstream-managed workflows. Use fixtures only when they are clearly labelled.
- After scaffolding, document the actual install, development, check and build commands in README and keep them synchronized with the package or project manifest. Do not report commands as successful unless they ran.
- Work in small steps. At handoff, list implemented flows, checks actually performed, remaining blockers, and any credentials or provider setup the owner must supply.
- Do not publish, spend money, contact customers, delete source data or run irreversible migrations without the project owner's authorization.

## Prerequisites
- A Python virtual environment, writable input/output directories and sufficient disk for both originals and outputs. Bind the service to localhost. Install FFmpeg and confirm codec support for the intended inputs. Authorized likeness/reference assets, one configured image/motion provider, provider keys and explicit per-render spending approval.
- Implementation components: Python, FastAPI and server-rendered HTML with HTMX for a local interface. SQLite for manifests and job state, with an explicit worker process and immutable source files. FFmpeg/ffprobe for explicit media operations and browser previews; never interpolate user filenames into shell commands.
- Scope boundary: No unauthorized impersonation, guaranteed motion fidelity or universal model availability.

## Stack and architecture
- Python, FastAPI and server-rendered HTML with HTMX for a local interface.
- SQLite for manifests and job state, with an explicit worker process and immutable source files.
- FFmpeg/ffprobe for explicit media operations and browser previews; never interpolate user filenames into shell commands.
- Domain model: consented likeness assets, source clips, approved first-frame revisions, provider jobs and render receipts

## Security and data integrity
- Bound file sizes and processing time, reject path traversal, and use argument arrays for subprocesses. Treat imported text as data and redact confidential source content from logs.
- Correctness boundary: Record permission to use the person's likeness; never auto-submit a paid render or present synthetic footage as authentic evidence.
- Save a job manifest with input hash, parameters and state. Write to temporary outputs, then atomically finalize only successful results; resume unfinished jobs without replacing originals.
- Export sources, manifests and outputs with checksums. Keep failed-job diagnostics and allow retry into a new output path; restore the database and file directory together.

## Agent implementation rules
- Project rule — data model: consented likeness assets, source clips, approved first-frame revisions, provider jobs and render receipts
- Project rule — preserve this invariant: Record permission to use the person's likeness; never auto-submit a paid render or present synthetic footage as authentic evidence.
- Project rule — acceptance evidence: Reject the proposed first frame and make no motion-generation request; an unknown provider timeout is reconciled before another paid job.

## Optional agent skills and references
- Optional external skill: [modern-python](https://github.com/trailofbits/skills/blob/main/plugins/modern-python/skills/modern-python/SKILL.md) — Set up Python projects with pyproject.toml, dependency management, linting, typing and automated checks. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
- Optional external skill: [sharp-edges](https://github.com/trailofbits/skills/blob/main/plugins/sharp-edges/skills/sharp-edges/SKILL.md) — Review security-sensitive APIs and configuration for dangerous defaults and easy-to-misuse interfaces. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
- Optional external skill: [web-design-guidelines](https://github.com/vercel-labs/agent-skills/blob/main/skills/web-design-guidelines/SKILL.md) — Review web interfaces for accessibility, keyboard focus, forms, navigation and interaction quality. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.

Read the linked SKILL.md and its dependencies before adding a skill. Select only the skills matching this project's runtime and task; their documentation does not supply API access, credentials or approval to perform external actions. Pin the reviewed revision where the tool supports it. Follow the chosen agent's documented project-level installation mechanism.

## Distribution ideas
These are optional planning notes. Obtain the owner's approval before publishing or contacting anyone.
- Demonstrate the actual Ghostfeed-inspired workflow with owned or clearly labeled sample data: Prepare a short user-owned clip, generate an optional identity-consistent first frame from an authorized likeness and require approval before requesting a paid motion render. Burn reviewed hook text into a separate export.
- Publish a reproducible walkthrough with this observable result: Reject the proposed first frame and make no motion-generation request; an unknown provider timeout is reconciled before another paid job.
- Explain who can operate this scoped tool, its setup and ongoing costs, and these remaining product gaps: No unauthorized impersonation, guaranteed motion fidelity or universal model availability. Avoid guaranteed savings, performance scores or implied endorsement.

## Engineering roadmap
1. Phase 1 — Scope and fixtures. Implement this bounded workflow: Prepare a short user-owned clip, generate an optional identity-consistent first frame from an authorized likeness and require approval before requesting a paid motion render. Burn reviewed hook text into a separate export. Record prerequisites, select representative user-owned fixtures and document the unsupported features: No unauthorized impersonation, guaranteed motion fidelity or universal model availability.
2. Phase 2 — Durable model. Model consented likeness assets, source clips, approved first-frame revisions, provider jobs and render receipts Add migrations or a versioned document format, explicit validation, stable IDs and a visible import-error report. Preserve this rule: Record permission to use the person's likeness; never auto-submit a paid render or present synthetic footage as authentic evidence.
3. Phase 3 — Complete the first useful path. Implement the workflow's input, review and output interface, with clear controls and explicit empty/error states. Save a job manifest with input hash, parameters and state. Write to temporary outputs, then atomically finalize only successful results; resume unfinished jobs without replacing originals.
4. Phase 4 — Permissions and integration failure. Bound file sizes and processing time, reject path traversal, and use argument arrays for subprocesses. Treat imported text as data and redact confidential source content from logs. Request integration credentials and permissions only for the enabled feature; show a disconnected state instead of mock results.
5. Phase 5 — Portable handoff. Export sources, manifests and outputs with checksums. Keep failed-job diagnostics and allow retry into a new output path; restore the database and file directory together. Include setup, operating limits, fixture walkthrough and shutdown/restart instructions in the README.
6. Phase 6 — Acceptance scenarios. Reject the proposed first frame and make no motion-generation request; an unknown provider timeout is reconciled before another paid job. Repeat the workflow after restart and with a denied permission or unavailable dependency; show recoverable failure rather than a success placeholder.

## Paid-product capabilities outside this build
- a template library scraped from TikTok and Instagram and scene-cut into ready 9:16 clips, instead of clips you source and trim yourself
- avatar consistency across renders · one stored reference photo re-fed into every call, plus the draft-and-approve flow that keeps a face you picked
- a director loop that writes the slideshow copy, searches for backgrounds, looks at the candidates and casts them, then rotates the deck into variants, plus a timeline editor with its own render workers
- 44 agent-API tools · the same workspace drivable from Claude, which is the part with no weekend equivalent
- a TikTok app that already cleared review, and a reaper that refunds most stalled renders instead of quietly eating them

## Implementation prompt
WORKING SLICE
Prepare a short user-owned clip, generate an optional identity-consistent first frame from an authorized likeness and require approval before requesting a paid motion render. Burn reviewed hook text into a separate export.

Build this scoped Ghostfeed-inspired workflow with a documented data model and visible failure states.

Architecture
- Python, FastAPI and server-rendered HTML with HTMX for a local interface.
- SQLite for manifests and job state, with an explicit worker process and immutable source files.
- FFmpeg/ffprobe for explicit media operations and browser previews; never interpolate user filenames into shell commands.

Prerequisites and limits
A Python virtual environment, writable input/output directories and sufficient disk for both originals and outputs. Bind the service to localhost. Install FFmpeg and confirm codec support for the intended inputs. Authorized likeness/reference assets, one configured image/motion provider, provider keys and explicit per-render spending approval.
Outside this release: No unauthorized impersonation, guaranteed motion fidelity or universal model availability.

Data model and correctness
consented likeness assets, source clips, approved first-frame revisions, provider jobs and render receipts
Invariant: Record permission to use the person's likeness; never auto-submit a paid render or present synthetic footage as authentic evidence.
Save a job manifest with input hash, parameters and state. Write to temporary outputs, then atomically finalize only successful results; resume unfinished jobs without replacing originals.

Security and privacy
Bound file sizes and processing time, reject path traversal, and use argument arrays for subprocesses. Treat imported text as data and redact confidential source content from logs.

Recovery and export
Export sources, manifests and outputs with checksums. Keep failed-job diagnostics and allow retry into a new output path; restore the database and file directory together.

Implementation order
1. Phase 1 — Scope and fixtures. Implement this bounded workflow: Prepare a short user-owned clip, generate an optional identity-consistent first frame from an authorized likeness and require approval before requesting a paid motion render. Burn reviewed hook text into a separate export. Record prerequisites, select representative user-owned fixtures and document the unsupported features: No unauthorized impersonation, guaranteed motion fidelity or universal model availability.
2. Phase 2 — Durable model. Model consented likeness assets, source clips, approved first-frame revisions, provider jobs and render receipts Add migrations or a versioned document format, explicit validation, stable IDs and a visible import-error report. Preserve this rule: Record permission to use the person's likeness; never auto-submit a paid render or present synthetic footage as authentic evidence.
3. Phase 3 — Complete the first useful path. Implement the workflow's input, review and output interface, with clear controls and explicit empty/error states. Save a job manifest with input hash, parameters and state. Write to temporary outputs, then atomically finalize only successful results; resume unfinished jobs without replacing originals.
4. Phase 4 — Permissions and integration failure. Bound file sizes and processing time, reject path traversal, and use argument arrays for subprocesses. Treat imported text as data and redact confidential source content from logs. Request integration credentials and permissions only for the enabled feature; show a disconnected state instead of mock results.
5. Phase 5 — Portable handoff. Export sources, manifests and outputs with checksums. Keep failed-job diagnostics and allow retry into a new output path; restore the database and file directory together. Include setup, operating limits, fixture walkthrough and shutdown/restart instructions in the README.
6. Phase 6 — Acceptance scenarios. Reject the proposed first frame and make no motion-generation request; an unknown provider timeout is reconciled before another paid job. Repeat the workflow after restart and with a denied permission or unavailable dependency; show recoverable failure rather than a success placeholder.

Acceptance
Reject the proposed first frame and make no motion-generation request; an unknown provider timeout is reconciled before another paid job.
Use real source data or clearly labeled fixtures. Explain unsupported input and provider failures; do not fabricate analytics, delivery receipts, accuracy claims or security guarantees.

Optional agent guidance
Optional external skill: [modern-python](https://github.com/trailofbits/skills/blob/main/plugins/modern-python/skills/modern-python/SKILL.md) — Set up Python projects with pyproject.toml, dependency management, linting, typing and automated checks. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
Optional external skill: [sharp-edges](https://github.com/trailofbits/skills/blob/main/plugins/sharp-edges/skills/sharp-edges/SKILL.md) — Review security-sensitive APIs and configuration for dangerous defaults and easy-to-misuse interfaces. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
Optional external skill: [web-design-guidelines](https://github.com/vercel-labs/agent-skills/blob/main/skills/web-design-guidelines/SKILL.md) — Review web interfaces for accessibility, keyboard focus, forms, navigation and interaction quality. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
Project rule — data model: consented likeness assets, source clips, approved first-frame revisions, provider jobs and render receipts
Project rule — preserve this invariant: Record permission to use the person's likeness; never auto-submit a paid render or present synthetic footage as authentic evidence.
Project rule — acceptance evidence: Reject the proposed first frame and make no motion-generation request; an unknown provider timeout is reconciled before another paid job.

## Completion evidence
Demonstrate the prompt's acceptance scenarios against the scoped workflow. Include setup from a clean checkout and failure recovery. Check persistence across restart and export/restore only for the state the prompt says to store; for memory-only tools, confirm that temporary content is discarded as specified. Record actual results and remaining limitations. A detailed plan alone does not establish a working replacement.