Give an agent a soft profile of likes and dislikes and take the must-haves off the form (wp-likes)

Add agents.like_text (migration 30) beside avoid_text, now shown as "What I don't like". The judge reads both under <likes> and <dislikes>, its career rubric judges against the prompts, the likes and the dislikes, and FORMAT 3 makes earlier judgements be made again. The form drops the must-haves box and no longer sends must_text, so a stored one is kept and still caps career in code.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
WolfandClaude Sonnet 5.5 committed 2026-10-03 22:48:46 +02:00
1 parent 82afa831f2
commit 8c6cbdf002
18 files changed
+144 -75

No files matched your search

+18 -12
View File
@@ -15,7 +15,7 @@ is guessed from ad text, the second is what the gates do.
The work is pairs, `(agent, role)`: an enabled agent with a CV, and a role
among its best ungated scores as deep and as low as its own `limits` say
(`judge_top`, `judge_min`) with some text, that has no judgement yet, or one
read from another CV, prompts, avoid or must text or role text than today's
read from another CV, prompts, like, avoid or must text or role text than today's
(`input_key`). A pair is named by the composite key `"<agent_id>:<uid>"`,
which is what `desk.modeltask` hands round as a uid, so `desk/` is untouched.
A batch holds one agent's pairs, since the profile is in the system prompt;
@@ -59,16 +59,18 @@ EXCERPT = 8000
#: The format of what the judge asks and stores, part of `input_key`: a
#: change to the prompt's contract makes every judgement be made again.
#: 2: the verdict on each numbered must-have (`musts`), the cap in code.
FORMAT = 2
#: 3: what the person likes (`like_text`) beside what they do not, and the
#: career rubric reading both.
FORMAT = 3
#: How many strengths and gaps are kept, and how long each may be.
MOST, LONGEST = 3, 300
SYSTEM = """You judge job postings for one person: how well each fits their CV and what they want.
Below the rules is the person: their CV, what they want and what excites
them, what drains them and environments to avoid, and their must-haves and
deal-breakers. Each item is one posting with a key, its title, company, place
Below the rules is the person: their CV, what they want, what they like and
what they do not like (the work, the workplace, situations), and their
must-haves and deal-breakers when they gave any. Each item is one posting with a key, its title, company, place
and text. Answer one judgement per item.
Score three dimensions, each 0 to 100:
@@ -79,11 +81,12 @@ Score three dimensions, each 0 to 100:
literal job title, and on its level. 80-100 direct experience in the same
domain and role type; 60-79 related, with transferable skills clear; 40-59
adjacent, the person would need to make the case; 0-39 unrelated.
- career: against what the person wants and what drains them. 80-100 strongly
- career: against what the person wants (the prompts), what they like and
what they do not like. 80-100 strongly
aligned with their direction, with clear growth; 60-79 a good role, partly
aligned; 40-59 a decent job that does not build toward their goals; 0-39 a
dead end or a step backwards. Judge whether the tasks would energise them,
not only whether they could do them.
dead end or a step backwards. Judge whether the tasks and the setting would
energise them, not only whether they could do them.
Their must-haves and deal-breakers are numbered M1, M2 and so on. Give a
verdict on every one of them, in `musts`, for each item:
@@ -173,10 +176,10 @@ def _text(row: Any) -> str:
def input_key(agent: Agent, text: str) -> str:
"""A hash of everything a judgement reads: the CV, the prompts, the
avoid and must texts, and the role's text as sent. A judgement stored
like, avoid and must texts, and the role's text as sent. A judgement stored
under another key is read again."""
blob = json.dumps([FORMAT, agent.cv_text, [[p.name, p.text] for p in agent.prompts],
agent.avoid_text, agent.must_text, text], ensure_ascii=False)
agent.like_text, agent.avoid_text, agent.must_text, text], ensure_ascii=False)
return hashlib.sha1(blob.encode()).hexdigest()[:16]
@@ -227,9 +230,12 @@ def system(agent: Agent) -> str:
for p in agent.prompts if p.text.strip())
if wants:
parts.append(f"\n<wants title=\"what they want and what excites them\">\n{wants}\n</wants>")
if agent.like_text.strip():
parts.append("\n<likes title=\"what they like: work, workplace, situations\">\n"
f"{agent.like_text.strip()}\n</likes>")
if agent.avoid_text.strip():
parts.append("\n<avoid title=\"what drains them and environments to avoid\">\n"
f"{agent.avoid_text.strip()}\n</avoid>")
parts.append("\n<dislikes title=\"what they do not like\">\n"
f"{agent.avoid_text.strip()}\n</dislikes>")
if musts := must_lines(agent.must_text):
lines = "\n".join(f"M{n}: {line}" for n, line in enumerate(musts, start=1))
parts.append(f"\n<must title=\"must-haves and deal-breakers\">\n{lines}\n</must>")
+7 -4
View File
@@ -1553,10 +1553,12 @@ class Agent(Model):
union over the enabled agents, and an agent with none means no board is
narrowed by words. They decide what is fetched, not what ranks.
`cv_text`, `cv_languages`, `avoid_text` and `must_text` are what
`judge` reads a role against: the CV, the languages it states with a
level, what drains the person, and their must-haves and deal-breakers
(soft, read by the model only; a hard rule is a gate). An agent without
`cv_text`, `cv_languages`, `like_text`, `avoid_text` and `must_text` are
what `judge` reads a role against: the CV, the languages it states with
a level, what the person likes and does not like (the work, the
workplace, situations; both feed the career score), and their
must-haves and deal-breakers (set through the API only, not on the
form; soft, read by the model only; a hard rule is a gate). An agent without
a CV is not judged, and one without languages flags none
(`language_flag`)."""
@@ -1569,6 +1571,7 @@ class Agent(Model):
fetch_terms: FetchTerms = []
cv_text: str = ""
cv_languages: CvLanguages = []
like_text: str = ""
avoid_text: str = ""
must_text: str = ""
enabled: bool = True
+14 -10
View File
@@ -20,12 +20,14 @@
*
* The CV fieldset holds what the judge stage reads a role against: the CV as
* text (typed, or read from a .md/.txt file by the browser), the languages it
* states as rows of a code and a level, and the avoid and must-have texts.
* The must-have text is one must-have per line, which the judge answers one
* verdict each; a broken one caps career at 39 in code (the hint in
* settings/agents.html says so).
* Every save sends all four, and the card's switch re-posts the row as the
* server gave it, so neither drops them.
* states as rows of a code and a level, and the two soft texts, what the
* person likes and what they do not like (the work, the workplace,
* situations), which the judge reads for the career score. The must-have text
* is no longer on the form: a save leaves `must_text` out, so the stored one
* is kept (it is set through the API, and still caps career at 39 in code
* when the judge reports one broken).
* Every save sends the CV, languages and the two texts, and the card's switch
* re-posts the row as the server gave it, so neither drops them.
*
* Every request goes through `api()` from app.js, so a failure is one toast
* carrying the server's own sentence.
@@ -502,12 +504,14 @@
}
// The CV on a card: whether the judge has one to read roles against, and
// the languages it states, as "da fluent, en native".
// the languages it states, as "da fluent, en native", and whether the
// likes and dislikes are written.
function cvLine(row) {
const langs = row.cv_languages || [];
const said = langs.length ? `Languages: ${langs.map((l) => `${l.language} ${l.level}`).join(", ")}.` : "No languages stated.";
const has = Boolean((row.cv_text || "").trim());
return el("p", "agent-terms", has ? `CV set. ${said}` : `No CV, so roles are not judged against one. ${said}`);
const soft = `Likes ${(row.like_text || "").trim() ? "set" : "not set"}, dislikes ${(row.avoid_text || "").trim() ? "set" : "not set"}.`;
return el("p", "agent-terms", `${has ? "CV set." : "No CV, so roles are not judged against one."} ${said} ${soft}`);
}
// A prompt as a row: its name and Remove on a line, its text under them.
@@ -646,8 +650,8 @@
// is cleared in the store.
const cvText = $("#a-cv-text", f);
cvText.value = row ? row.cv_text || "" : "";
$("#a-like-text", f).value = row ? row.like_text || "" : "";
$("#a-avoid-text", f).value = row ? row.avoid_text || "" : "";
$("#a-must-text", f).value = row ? row.must_text || "" : "";
const languages = [];
const dropLanguage = (node) => {
languages.splice(languages.findIndex((l) => l.node === node), 1);
@@ -898,8 +902,8 @@
fetch_terms: read.terms(),
cv_text: $("#a-cv-text", f).value,
cv_languages: cvLanguages,
like_text: $("#a-like-text", f).value,
avoid_text: $("#a-avoid-text", f).value,
must_text: $("#a-must-text", f).value,
// Saving an agent means it is wanted, so a new one starts on; the
// card's switch is where an existing one is switched.
enabled: row ? row.enabled : true,
+18 -4
View File
@@ -328,9 +328,10 @@ END;
-- as JSON: a key it lacks means the `Limits` default. fetch_terms is a JSON
-- list of words that narrow what big boards send (app/jobs/reach.py); empty
-- means no narrowing. cv_text, cv_languages (a JSON list of {language,
-- level}), avoid_text (what drains) and must_text (must-haves and
-- deal-breakers) are what judge reads a role against; an agent with no CV
-- is not judged.
-- level}), like_text (what they like: the work, the workplace, situations),
-- avoid_text (what they do not like) and must_text (must-haves and
-- deal-breakers, set through the API only) are what judge reads a role
-- against; an agent with no CV is not judged.
CREATE TABLE IF NOT EXISTS agents (
id INTEGER PRIMARY KEY,
name TEXT NOT NULL UNIQUE,
@@ -343,7 +344,8 @@ CREATE TABLE IF NOT EXISTS agents (
cv_text TEXT NOT NULL DEFAULT '',
cv_languages TEXT NOT NULL DEFAULT '[]',
avoid_text TEXT NOT NULL DEFAULT '',
must_text TEXT NOT NULL DEFAULT ''
must_text TEXT NOT NULL DEFAULT '',
like_text TEXT NOT NULL DEFAULT ''
);
-- What an agent looks for, as several named texts in order; a role's fit is
@@ -707,6 +709,10 @@ UPDATE roles SET seniority = COALESCE(NULLIF(seniority_model, ''),
# keep their numbers; the judge reads them again (judge.FORMAT).
(29, "role_judgements.musts and broken: the verdict on each must-have, and how many are "
"broken", lambda conn: _add_musts(conn)),
# A Python step: it adds the column only where it is missing. The judge
# reads every role again (judge.FORMAT).
(30, "agents.like_text: what the person likes, beside what they do not like",
lambda conn: _add_like_text(conn)),
)
#: What a file created from `SCHEMA` today is already at.
@@ -1047,6 +1053,14 @@ def _add_musts(conn: sqlite3.Connection) -> None:
conn.execute(f"ALTER TABLE role_judgements ADD COLUMN {column} {ddl}")
def _add_like_text(conn: sqlite3.Connection) -> None:
"""Migration 30's step: `agents.like_text`, empty on every agent. Runs
twice without harm."""
columns = {r[1] for r in conn.execute("PRAGMA table_info(agents)")}
if "like_text" not in columns:
conn.execute("ALTER TABLE agents ADD COLUMN like_text TEXT NOT NULL DEFAULT ''")
def _add_all_roles(conn: sqlite3.Connection) -> None:
"""Migration 21's step. The column; the tag for every company that is a
favourite of any agent, since favourite no longer widens what a pull
+10 -6
View File
@@ -2093,13 +2093,14 @@ class Store:
cv_text: str | None = None,
cv_languages: Sequence[CvLanguage] | None = None,
avoid_text: str | None = None,
must_text: str | None = None) -> Agent:
must_text: str | None = None,
like_text: str | None = None) -> Agent:
"""Create an agent, or with `agent_id` replace that one's fields and
prompts. Without `prompts`, a non-empty `intent` is its one prompt,
named after the agent. Without `fetch_terms` an update keeps the
stored ones and a new agent has none, so a caller that does not know
them (the CLI's prompt edit) loses none; `cv_text`, `cv_languages`,
`avoid_text` and `must_text` are kept the same way. A prompt whose text is
`like_text`, `avoid_text` and `must_text` are kept the same way. A prompt whose text is
unchanged keeps its cached vector, whatever its name or place, so an
edit re-embeds only the prompts it changed. A save gives every role
`judge` gave up on for the agent its tries back, since what it is
@@ -2120,16 +2121,18 @@ class Store:
(limits or Limits()).model_dump_json(), int(enabled), terms,
None if cv_text is None else cv_text.strip(), spoken,
None if avoid_text is None else avoid_text.strip(),
None if must_text is None else must_text.strip())
None if must_text is None else must_text.strip(),
None if like_text is None else like_text.strip())
try:
with self.transaction() as conn:
if agent_id is None:
agent_id = conn.execute(
"""INSERT INTO agents (name, gates, weights, limits, enabled,
fetch_terms, cv_text, cv_languages, avoid_text,
must_text, created_at)
must_text, like_text, created_at)
VALUES (?,?,?,?,?,COALESCE(?, '[]'),COALESCE(?, ''),
COALESCE(?, '[]'),COALESCE(?, ''),COALESCE(?, ''),?)""",
COALESCE(?, '[]'),COALESCE(?, ''),COALESCE(?, ''),
COALESCE(?, ''),?)""",
(*fields, self.now())).lastrowid
elif not conn.execute(
"""UPDATE agents SET name=?, gates=?, weights=?, limits=?, enabled=?,
@@ -2137,7 +2140,8 @@ class Store:
cv_text=COALESCE(?, cv_text),
cv_languages=COALESCE(?, cv_languages),
avoid_text=COALESCE(?, avoid_text),
must_text=COALESCE(?, must_text)
must_text=COALESCE(?, must_text),
like_text=COALESCE(?, like_text)
WHERE id=?""",
(*fields, agent_id)).rowcount:
raise KeyError(agent_id)
+4 -4
View File
@@ -93,11 +93,11 @@
<button type="button" id="a-add-language" class="secondary">Add language</button>
<button type="button" id="a-suggest-languages" class="secondary">Suggest languages</button>
<span class="hint">A two-letter code such as <code>en</code> or <code>da</code>, and how well. Roles that ask for a language you lack, or a level above yours, get a warning on Roles; a role that names one of yours as a plus gets a star. With none, no role is flagged. Suggest languages reads them from the CV with one model call and replaces the rows above; nothing is saved until Save.</span>
<label>Avoid / drains me
<label>What I like
<textarea id="a-like-text" rows="4" placeholder="Building a team from scratch, close work with the product side, small companies, a say in the roadmap"></textarea></label>
<label>What I don't like
<textarea id="a-avoid-text" rows="4" placeholder="Constant on-call, sales targets, large consultancies, open-plan offices"></textarea></label>
<label>Must-haves and deal-breakers
<textarea id="a-must-text" rows="4" placeholder="Permanent contract&#10;A say in the roadmap&#10;At most one day a week in the office"></textarea></label>
<span class="hint">One must-have per line. The judge reports for each whether a posting meets it, clearly breaks it or does not say; a broken one caps the role's career score at 39 in code, whatever the model scored. Rules that must hold go under Where or Other rules instead.</span>
<span class="hint">The work, the workplace, situations; read by the CV match for the career score. Rules that must hold go under Where or Other rules instead.</span>
</fieldset>
<fieldset class="weights">
<legend>How much each counts</legend>
+4 -2
View File
@@ -1205,7 +1205,7 @@ class AgentBody(BaseModel):
it is stored as one prompt named after the agent. `fetch_terms` narrow
what big boards send (`app/jobs/reach.py`); `[]` narrows nothing, and
leaving the key out keeps an existing agent's. `cv_text`,
`cv_languages`, `avoid_text` and `must_text` are what `judge` reads a
`cv_languages`, `like_text`, `avoid_text` and `must_text` are what `judge` reads a
role against (`app/jobs/judge.py`), kept the same way when left out or
null."""
@@ -1219,6 +1219,7 @@ class AgentBody(BaseModel):
fetch_terms: FetchTerms | None = None
cv_text: str | None = None
cv_languages: CvLanguages | None = None
like_text: str | None = None
avoid_text: str | None = None
must_text: str | None = None
enabled: bool = True
@@ -1260,7 +1261,8 @@ def save_agent(body: AgentBody, store: StoreDep, queue: QueueDep) -> Agent:
body.enabled, agent_id=body.id, limits=body.limits,
prompts=body.prompts, fetch_terms=body.fetch_terms,
cv_text=body.cv_text, cv_languages=body.cv_languages,
avoid_text=body.avoid_text, must_text=body.must_text)
like_text=body.like_text, avoid_text=body.avoid_text,
must_text=body.must_text)
except ValueError as exc:
raise HTTPException(409, str(exc)) from exc
except KeyError: