# Reusable Prompt — Thread Artefact Catalogue **Purpose:** capture, in a single archived `.md` file, a verifiable record of everything an AI thread produced — files, conversational outputs, and referenced material — so the work can be audited later. **Design principle:** the catalogue must be honest about what it can and cannot verify. The discipline is Chapter-Six-of-Mind-the-Gap discipline: do not let a confident-looking entry sit beside an unreconciled one without marking the difference. --- ## The prompt — paste this into any thread Produce a comprehensive catalogue, as an `.md` file saved to `/mnt/user-data/outputs/`, of every artefact this thread has produced or referenced. Cover three categories, each clearly demarcated: **Category A — Files saved to `/mnt/user-data/outputs/` in this thread.** These are your direct deliverables. For each, list: 1. Sequential reference number (A1, A2, A3…) 2. Filename exactly as saved 3. File `mtime` (the disk timestamp of creation) 4. The document's internal date if it differs from the `mtime` 5. Word count (extracted via the appropriate tool) or file size in bytes if word count is not extractable 6. Brief description in one or two sentences — what the artefact is and what it does 7. **Verification status:** `[VERIFIED ON DISK]` or `[NOT FOUND ON DISK]` — based on a direct directory listing performed at the time the catalogue is written **Category B — Artefacts referenced from earlier scope (transcript summary, pre-thread work, or prior sessions) but not directly verifiable from this thread's outputs.** For each: 1. Sequential reference number (B1, B2, B3…) 2. Filename as referenced (with `[as referenced]` flag if you are not certain of the exact filename) 3. Date as referenced 4. Brief description of what the artefact is 5. **Source of the reference** — e.g. "named in transcript summary," "referenced in conversation" 6. **Verification status:** `[REFERENCED — NOT DIRECTLY VERIFIED]` **Category C — Substantial conversational outputs that did not become files but were significant pieces of work.** For each: 1. Sequential reference number (C1, C2, C3…) 2. Approximate date or position in the thread 3. Description of what was produced — the topic, the structural work, the reasoning piece 4. Approximate word count if known, otherwise marked `[in-conversation]` 5. Why it is being catalogued — what makes it a piece of work worth preserving rather than ordinary conversational exchange **Thread metadata — list at the top of the catalogue:** - **Your identifier in this thread** (if a name, persona, or working identity has been used — e.g. "Book Man," "Jose," or whatever the thread has been called). If none, mark `[no identifier used]`. - **Earliest verifiable artefact timestamp** in Category A - **Latest verifiable artefact timestamp** in Category A - **Current date** at time of writing the catalogue - **Scope start date** — the date the earliest referenced or produced artefact in *any* category dates to, marked with its source - **Approximate thread duration** (earliest scope to current date) **Validation discipline:** - Before listing each Category A entry, perform a `ls -la /mnt/user-data/outputs/` (or equivalent directory listing) and check the file is actually present. If it is missing, mark `[NOT FOUND ON DISK]` rather than silently omitting or assuming. - For Category A, the `mtime` is the canonical creation timestamp. The document's internal date is a *claim made inside the document* and may differ — list both where they differ, and note the divergence in the description. - For Category B, be explicit that the timestamps and filenames are *as referenced*, not as verified. - For Category C, do not inflate. Conversational exchanges that were not substantial pieces of work should not be catalogued here. Use judgement: if it took meaningful reasoning, produced a structural argument, or would be useful to retrieve later, include it. If it was ordinary dialogue, do not. **Honest closing section:** End the catalogue with a short section titled *"Honest reconciliation notes."* In this section, name: - Any Category A files where the `mtime` and the document's internal date diverge meaningfully - Any Category B entries you are genuinely uncertain about - Anything you suspect may exist but cannot verify - Anything in your context that suggests files were produced but have not appeared in your directory listing **Formatting:** - Save as `.md` to `/mnt/user-data/outputs/` - Filename should include the date and a thread-identifying suffix where possible (e.g. `Thread-Artefact-Catalogue-2026-06-09-BookMan.md`) - Use headers, tables where useful, and inline markers (`[VERIFIED ON DISK]`, `[REFERENCED]`, `[in-conversation]`) consistently so the document is scannable **Refusal to smooth:** If you cannot complete any category — e.g. the directory listing fails, you have no transcript summary visible, you do not know your thread identifier — say so explicitly in the relevant section. Do not fill in confident-looking entries to cover the gap. The reader will rely on this catalogue as audit material. An honest *"I cannot see this"* is more useful than a smoothed-over reconstruction. --- ## Notes on using this prompt across multiple threads - The prompt is thread-agnostic. Paste it as a complete message in any AI thread where you want a catalogue of what was produced. - Each thread's catalogue will be honest within its own context — it will not be able to verify artefacts from other threads. - To build a complete cross-thread archive, run the prompt in each thread separately and consolidate the resulting `.md` files locally. - Where threads have used persona names or working identities (Book Man, Jose, etc.), the catalogues will be self-labelled by that identity, making consolidation straightforward.