fix: stop publishing prospect data; correct stale profile guidance

.env.example

@@ -11,6 +11,17 @@ OPENROUTER_API_KEY=
 # Optional: Hunter.io email-name enrichment (skipped when absent)
 #HUNTER_API_KEY=
 
+# --- Offer profiles (feature 008) ---
+# Default profile, so --profile can be omitted. Without either, an interactive
+# run lists the available profiles and asks; a non-interactive one fails.
+#PROSPECTOR_PROFILE=duct-cleaning
+# Extra profile directory, searched BEFORE ./profiles/ and before the profiles
+# bundled with the package. Your own directory always wins.
+#PROSPECTOR_PROFILES=/path/to/my/profiles
+
+# Optional: vault output folder (default: Vault/Outreach). --vault overrides it.
+#PROSPECTOR_VAULT=Vault/Outreach
+
 # --- Approved-send (features 003/004) ---
 # Provider: gmail (Gmail API, default) or smtp (authenticated SMTP, e.g. Zoho).
 #PROSPECTOR_SEND_PROVIDER=smtp

.github/ISSUE_TEMPLATE/bug_report.md

@@ -12,7 +12,7 @@ labels: bug
 ## Command you ran
 
 ```bash
-# e.g. prospector run companies.csv --limit 3
+# e.g. prospector run companies.csv --profile duct-cleaning --limit 3
 ```
 
 ## Expected vs actual

.gitignore

@@ -17,6 +17,14 @@ send_ledger.jsonl
 Vault/
 samples/
 
+# Prospect data must never be committed: these files hold real businesses'
+# names, addresses and websites. `candidates.csv` is the default --out of
+# `prospector source`, so it appears in the working tree on a normal run.
+candidates.csv
+*candidates*.csv
+dm_ledger.jsonl
+*.local.csv
+
 # Internal development method — not part of the public project.
 # These stay on the maintainer's machine and are intentionally untracked.
 .specify/

CONTRIBUTING.md

@@ -132,6 +132,11 @@ fixes should include a test that fails before the fix and passes after.
   frameworks or agent/orchestration machinery.
 - Never hardcode secrets. Credentials come from the gitignored `.env`; never log
   or commit them.
+- **Never commit prospect data.** Real company names, addresses, websites and
+  scraped emails belong to third parties and must stay local. The vault,
+  `candidates.csv` (the default `--out` of `prospector source`), the send ledger
+  and `samples/` are all gitignored for this reason. Use fictional data in tests,
+  fixtures, issues and PR output.
 
 ## Commit and PR conventions
 

README.md

@@ -183,6 +183,7 @@ Secrets are loaded from the gitignored `.env` file.
 | `OPENROUTER_MODEL` | No | Defaults to `anthropic/claude-sonnet-4.5`. |
 | `PROSPECTOR_PROFILE` | No | Default profile name, so `--profile` can be omitted. |
 | `PROSPECTOR_PROFILES` | No | Extra profile directory, searched before `./profiles/`. |
+| `PROSPECTOR_VAULT` | No | Vault output folder; defaults to `Vault/Outreach`. `--vault` overrides it. |
 | `GOOGLE_PLACES_API_KEY` | For `source` | Required for discovery. During `run`, its absence enables the DuckDuckGo fallback. |
 | `HUNTER_API_KEY` | No | Enables email-name enrichment at medium confidence. |
 | `PROSPECTOR_SEND_PROVIDER` | No | `gmail` (default) or `smtp`. |
@@ -197,6 +198,8 @@ Secrets are loaded from the gitignored `.env` file.
 | `PROSPECTOR_SEND_CAPS` | No | Weekly cap ramp; defaults to `15,30,60,100`. |
 | `PROSPECTOR_SEND_DELAY` | No | Delay range in seconds; defaults to `30,90`. |
 | `PROSPECTOR_LEDGER` | No | Ledger path; defaults to `send_ledger.jsonl`. |
+| `PROSPECTOR_GMAIL_CLIENT` | No | Gmail OAuth client secret; defaults to `secrets/gmail_client_secret.json`. |
+| `PROSPECTOR_GMAIL_TOKEN` | No | Stored Gmail token; defaults to `secrets/gmail_token.json`. |
 
 Gmail OAuth files live under `secrets/`; the send ledger remains local. Both
 locations are excluded from version control.

prospector/profiles/duct-cleaning/CONSTRAINTS.md

@@ -41,8 +41,8 @@ program owns. If you write a name yourself, the draft is rejected.
 `cites` is a list of identifiers, and **it must never be empty**.
 
 You are given an evidence catalogue for this company. Each entry has an `id`
-like `about_page_1`, `hook_source_1`, or `fb_link_1`. Those ids are what you
-cite.
+like `about_page_1`, `hook_source_1`, or `email_published_1`. Those ids are what
+you cite. Cite only ids that appear in the catalogue you were given.
 
 - **A block that says something about the prospect** cites the evidence
   id(s) that support it. If you write "you have been serving Dallas for 22
@@ -85,18 +85,23 @@ The word "your" is what turns one into the other:
 | "when someone messages **your page** at 9pm" | "when someone messages a business at 9pm" |
 | "It watches **your inbox**" | "It watches the page inbox" |
 
-Banned outright unless the evidence catalogue contains an `fb_*` record, you
-cite it in that same block, **and** you were told the signal is strong:
+**Banned outright — there is no evidence that makes these acceptable:**
 "your facebook page", "your fb page", "your page", "your inbox",
 "your messenger", "your dms", "your direct messages", "messages your page".
 
+There used to be an exception here for a cited `fb_*` evidence record on a
+strong channel signal. **That exception is gone.** The tool no longer researches,
+scores, or records anything about a prospect's Facebook usage — there is no
+`fb_*` record to cite and no signal to be told about — so a possessive channel
+phrase can never be justified. A validator rejects these unconditionally.
+
 A page appearing in search results is not proof they read it. Say what the tool
 does. Never say what they own.
 
-- **Never assert they use Facebook** unless you were given `fb_*` evidence and
-  you cite it. Describing what the *product* does with a Facebook inbox is a
-  fact about the product and is always fine. Saying *they* are active on
-  Facebook requires evidence.
+- **Never assert they use Facebook.** Not conditionally, not with a hedge.
+  Describing what the *product* does with a Facebook inbox is a fact about the
+  product and is always fine. Saying *they* are on Facebook is a claim about the
+  prospect, and nothing in the evidence catalogue can support it.
 - Never invent a problem, a metric, a compliment, or a number.
 - Never guarantee bookings, revenue, or replies.
 

prospector/profiles/duct-cleaning/skills/write-cold-email.md

@@ -30,8 +30,10 @@ before you write anything. Different evidence supports different openings:
   and repeat customers, and a dropped message costs more.
 - A service area or city supports an opening about their market.
 - An about/team page naming the owner supports addressing what they built.
-- A Facebook page link supports mentioning that channel, but only as a fact
-  about where messages arrive, never as a claim about how active they are.
+
+Nothing in the catalogue describes the prospect's own marketing channels, so no
+opening can be built on one. Do not reach for Facebook, Messenger, ads, or "where
+your leads come from" — you have not been told, and guessing is a fabrication.
 
 **Never open with:**
 - "I hope this email finds you well."