Ephemeral Testing: Task Descriptions and Agent Prompts

Companion to ephemeral_testing_targets.md. Each task gives a goal, the inputs the agent may use, and a concise prompt to hand to a fresh coding agent.

Ground rules (apply to every task)

• Work in a new, empty directory outside the repo. Nothing built is committed here.

• Allowed inputs: the docs and artifacts listed per task. Do not read src/ unless the task says so.

• Keep a FINDINGS.md in the working directory. Log every ambiguity, question, workaround, surprising default and unhelpful error message, with a one-line note on what doc, error message or API change would have prevented it.

• Stop after the deliverable works or after being blocked for a meaningful time; record why.

• Triage findings afterwards as: doc fix, error-message fix, or API change.

Common prompt preamble

You are testing a codebase by building a throwaway tool on top of it. Use only

the inputs listed below. Do not read implementation source unless told to.

Maintain FINDINGS.md: log every point of confusion, workaround, surprising

behavior or poor error message, and what would have prevented it. Findings

about the project are the real deliverable; the tool is disposable.

Task 1: Read-only interface from a database dump

Goal: Build a read-only web or CLI browser over a SQLite dump of the forum.

Inputs: a dump of the read model, queries/sqlite catalog, docs/specs/read_model_schema_v1.md, README.md.

Deliverable: threads list, thread view and search, with no write paths.

Prompt:

Build a read-only browser (small web app or CLI) over the attached SQLite dump.

It must list threads, show a thread with its posts in order, and support basic

search. Infer the schema from the dump and docs only. Open the DB read-only.

Log in FINDINGS.md anything you had to guess about the schema, ordering,

visibility or moderation fields.

Task 2: Independent canonical-record parser and verifier

Goal: Reimplement record parsing and signature verification from the spec.

Inputs: the record specs in docs/specs/ (canonical_post_record_v1.md, identity_bootstrap_record_v1.md, public_key_storage_v1.md, and the other *_record_v1.md files), docs/architecture/public_architecture_and_trust.md, and a sample set of raw records with their detached signatures and public keys (no parser source).

Deliverable: a parser and verifier in Python or JS, plus a report comparing its results with the repo's scripts/audit_post_signatures.php output on the samples.

Prompt:

Implement a parser and signature verifier for the canonical records in the

sample directory, in Python, using only the specs in docs/specs/ and the trust

architecture doc. Do not read PHP source. Expect that exactly what is signed and

how verification works may be underspecified; log each such gap. Output one line per record: id, valid/invalid, reason. Log every

place the spec was ambiguous or silent, and every disagreement with the

reference results.

Task 3: Third-party posting client

Goal: Create a key, sign a post and submit it to a local dev instance.

Inputs: public docs, a running dev instance (scripts/dev_server.php).

Deliverable: a script that registers or identifies a key, signs, posts, and reads the post back.

Prompt:

Write a minimal client that generates an OpenPGP key, signs a post, submits it

to the dev instance at the given URL, and fetches it back. Use only public

docs. Log every rejection message you hit and whether it told you what to fix.

Task 4: Alternate static renderer

Goal: Render the archive with a different layout from the published data.

Inputs: a repository archive or state/static_html snapshot, docs.

Deliverable: a single-page archive viewer or feed generated from the data.

Prompt:

Generate a self-contained static archive viewer (single HTML file or small

folder) from the attached repository archive, with a different layout than the

existing site. Use only the archive and docs. Log what you had to reverse-engineer

about file layout, naming and cross-references.

Task 5: Foreign-format importer

Goal: Import a small non-forum dataset (e.g. mailing-list mbox) as posts.

Inputs: the qdb_archive_import_* scripts as the documented precedent, docs.

Deliverable: an importer that emits records the system accepts.

Prompt:

Using the qdb archive import as the model, write an importer that converts the

attached mbox into valid post records. Validate the output with the repo's

tooling. Log which parts of the qdb importer were reusable and which were

hard-wired to qdb.

Task 6: Agent-reply provider swap

Goal: Add an alternate reply provider behind the existing contract.

Inputs: provider interface docs, scripts/test_agent_reply_provider.php, tests. Reading the interface file is allowed; reading the Anthropic provider is not.

Deliverable: a stub or alternate provider that passes the provider tests.

Prompt:

Add a new agent-reply provider (a deterministic fake is fine) that satisfies the

existing provider contract and passes the provider tests. Do not read the

Anthropic provider implementation. Log every assumption the contract made

implicitly.

Task 7: Operator status report

Goal: Produce a health summary using only the operator scripts.

Inputs: scripts/*.php and their --help output, docs/reference/v3_cli.md, docs/runbooks/, README.md.

Deliverable: a script or markdown report showing queue state, recent errors and index freshness.

Prompt:

Build an operator status report using only the CLI scripts in scripts/ and

their help output. Do not read script source. Log which questions you could not

answer from the CLI and which flags or outputs were unclear.

Task 8: Blind deploy

Goal: Stand up a fresh instance from the docs alone.

Inputs: README.md, CONTRIBUTING.md, a clean clone.

Deliverable: a running local instance with one signed post visible.

Prompt:

From a clean clone, get a local instance running and create one signed post,

following only README.md and CONTRIBUTING.md. Record each step, each

command that failed, and each missing prerequisite.

Suggested order

1. Tasks 1 and 2 first (data layer and spec).

2. Then 3 and 8 (external contract and onboarding).

3. Tasks 4 to 7 as time allows.

Findings triage template