jobforge 0.1.0rc2__tar.gz

jobforge-0.1.0rc2/LICENSE

@@ -0,0 +1,75 @@
+Required Notice: Copyright 2026 Jesse Anderson and Mike Wilson
+
+# PolyForm Noncommercial License 1.0.0
+
+<https://polyformproject.org/licenses/noncommercial/1.0.0>
+
+## Acceptance
+
+In order to get any license under these terms, you must agree to them as both strict obligations and conditions to all your licenses.
+
+## Copyright License
+
+The licensor grants you a copyright license for the software to do everything you might do with the software that would otherwise infringe the licensor's copyright in it for any permitted purpose.  However, you may only distribute the software according to [Distribution License](#distribution-license) and make changes or new works based on the software according to [Changes and New Works License](#changes-and-new-works-license).
+
+## Distribution License
+
+The licensor grants you an additional copyright license to distribute copies of the software.  Your license to distribute covers distributing the software with changes and new works permitted by [Changes and New Works License](#changes-and-new-works-license).
+
+## Notices
+
+You must ensure that anyone who gets a copy of any part of the software from you also gets a copy of these terms or the URL for them above, as well as copies of any plain-text lines beginning with `Required Notice:` that the licensor provided with the software.  For example:
+
+> Required Notice: Copyright Yoyodyne, Inc. (http://example.com)
+
+## Changes and New Works License
+
+The licensor grants you an additional copyright license to make changes and new works based on the software for any permitted purpose.
+
+## Patent License
+
+The licensor grants you a patent license for the software that covers patent claims the licensor can license, or becomes able to license, that you would infringe by using the software.
+
+## Noncommercial Purposes
+
+Any noncommercial purpose is a permitted purpose.
+
+## Personal Uses
+
+Personal use for research, experiment, and testing for the benefit of public knowledge, personal study, private entertainment, hobby projects, amateur pursuits, or religious observance, without any anticipated commercial application, is use for a permitted purpose.
+
+## Noncommercial Organizations
+
+Use by any charitable organization, educational institution, public research organization, public safety or health organization, environmental protection organization, or government institution is use for a permitted purpose regardless of the source of funding or obligations resulting from the funding.
+
+## Fair Use
+
+You may have "fair use" rights for the software under the law. These terms do not limit them.
+
+## No Other Rights
+
+These terms do not allow you to sublicense or transfer any of your licenses to anyone else, or prevent the licensor from granting licenses to anyone else.  These terms do not imply any other licenses.
+
+## Patent Defense
+
+If you make any written claim that the software infringes or contributes to infringement of any patent, your patent license for the software granted under these terms ends immediately. If your company makes such a claim, your patent license ends immediately for work on behalf of your company.
+
+## Violations
+
+The first time you are notified in writing that you have violated any of these terms, or done anything with the software not covered by your licenses, your licenses can nonetheless continue if you come into full compliance with these terms, and take practical steps to correct past violations, within 32 days of receiving notice.  Otherwise, all your licenses end immediately.
+
+## No Liability
+
+***As far as the law allows, the software comes as is, without any warranty or condition, and the licensor will not be liable to you for any damages arising out of these terms or the use or nature of the software, under any kind of legal claim.***
+
+## Definitions
+
+The **licensor** is the individual or entity offering these terms, and the **software** is the software the licensor makes available under these terms.
+
+**You** refers to the individual or entity agreeing to these terms.
+
+**Your company** is any legal entity, sole proprietorship, or other kind of organization that you work for, plus all organizations that have control over, are under the control of, or are under common control with that organization.  **Control** means ownership of substantially all the assets of an entity, or the power to direct its management and policies by vote, contract, or otherwise.  Control can be direct or indirect.
+
+**Your licenses** are all the licenses granted to you for the software under these terms.
+
+**Use** means anything you do with the software requiring one of your licenses.

jobforge-0.1.0rc2/PKG-INFO

@@ -0,0 +1,183 @@
+Metadata-Version: 2.4
+Name: jobforge
+Version: 0.1.0rc2
+Summary: Tailored resumes and cover letters from your real work history, run locally, with every fact checked.
+Author: Jesse Anderson, Mike Wilson
+License-Expression: LicenseRef-PolyForm-Noncommercial-1.0.0
+Project-URL: Homepage, https://github.com/jesse-anderson/jobforge
+Project-URL: Source, https://github.com/jesse-anderson/jobforge
+Project-URL: Issues, https://github.com/jesse-anderson/jobforge/issues
+Project-URL: Security, https://github.com/jesse-anderson/jobforge/security/policy
+Classifier: Programming Language :: Python :: 3
+Classifier: Operating System :: Microsoft :: Windows
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Environment :: Console
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: dev
+Requires-Dist: pytest>=8; extra == "dev"
+Requires-Dist: coverage>=7; extra == "dev"
+Dynamic: license-file
+
+# Jobforge
+
+[![tests](https://github.com/jesse-anderson/jobforge/actions/workflows/tests.yml/badge.svg)](https://github.com/jesse-anderson/jobforge/actions/workflows/tests.yml)
+[![coverage](https://img.shields.io/endpoint?url=https://jesse-anderson.github.io/jobforge/coverage-badge.json)](https://github.com/jesse-anderson/jobforge/actions/workflows/tests.yml)
+[![python](https://img.shields.io/badge/python-3.11%2B-blue)](https://www.python.org/downloads/)
+[![license](https://img.shields.io/badge/license-PolyForm%20Noncommercial%201.0.0-blue)](https://github.com/jesse-anderson/jobforge/blob/main/LICENSE)
+[![platforms](https://img.shields.io/badge/platforms-Windows%20%7C%20Linux-lightgrey)](https://github.com/jesse-anderson/jobforge#what-you-need)
+[![dependencies](https://img.shields.io/badge/runtime%20dependencies-none-brightgreen)](https://github.com/jesse-anderson/jobforge/blob/main/pyproject.toml)
+
+Turn your real work history into a resume and cover letter tailored to one
+job posting, on your own computer, for free, without the tool ever stating a
+fact you did not give it.
+
+**Status: pre-release.** It works end to end, but it is not on PyPI yet.
+
+- **Free and local.** Nothing runs on a server. Your AI coding agent (Claude
+  Code first; Codex, Gemini CLI and others also work) does the writing, and
+  Jobforge does everything else.
+- **No invented facts.** Every number, employer, tool, and credential in your
+  documents is checked against facts you confirmed. If your agent invents
+  one, Jobforge refuses to write the document.
+- **Built for applicant tracking systems.** It writes `.docx` files with no
+  tables, columns, or hidden characters, then reads them back to prove what a
+  parser will see.
+
+Source-available under [PolyForm Noncommercial 1.0.0](https://github.com/jesse-anderson/jobforge/blob/main/LICENSE): free for personal use
+and for nonprofits, schools, and public institutions. Not for commercial use.
+
+## What you need
+
+- Windows or Linux (macOS is untested).
+- Python 3.11 or newer. Check with `python3 --version` on Linux, or
+  `python --version` on Windows.
+- An AI coding agent that runs in a terminal, such as Claude Code.
+
+## Set up (once)
+
+On Linux, install with pipx. Ubuntu, Debian, and others block `pip` for the
+system Python, and pipx is how they want apps like this one installed:
+
+```
+sudo apt install pipx
+pipx ensurepath                 # then open a new terminal
+pipx install jobforge           # or: git clone, then `pipx install .`
+jobforge init ~/jobsearch
+```
+
+On Windows:
+
+```
+pip install jobforge            # or: git clone, then `pip install .`
+jobforge init ~/jobsearch
+```
+
+That makes a folder with everything Jobforge needs:
+
+```
+jobsearch/
+  CLAUDE.md, AGENTS.md   instructions for your agent (generated)
+  profile/profile.toml   your name and contact details
+  profile/atoms/         your confirmed facts
+  profile/pending/       facts your agent drafted, waiting for you
+  intake/                put your old resume here
+  jobs/                  one folder per job you apply to
+```
+
+1. Copy your old resume (and any cover letters) into `intake/`.
+2. Open your agent in the `jobsearch` folder and say **"set me up"**. It reads
+   your resume, fills in your profile, and drafts your facts into
+   `profile/pending/`. Where your resume is unclear, it asks.
+3. Run **`jobforge confirm`**. For each drafted fact, you choose keep, drop,
+   or skip. Only facts you keep are ever used.
+
+To remove Jobforge later, `pipx uninstall jobforge` (Linux) or
+`pip uninstall jobforge` (Windows). Your `jobsearch` folder stays as it is.
+
+## Apply to a job
+
+Tell your agent: **"apply to this job: <link>"** (or paste the posting). It
+will:
+
+1. `jobforge prepare --from-url <link>` to save the posting,
+2. write a plan for which facts to use and how to word them,
+3. `jobforge check <job>` until the plan passes,
+4. `jobforge apply <job>`, which writes `Your_Name_Resume.docx` and
+   `Your_Name_Cover_Letter.docx` into the job's folder.
+
+Read both documents, then upload the `.docx` files yourself. `cover_letter.md`
+is a working copy for review; never send it.
+
+When you have applied: `jobforge applied --mark <job> --status applied`.
+Record later news the same way (`screen`, `interview`, `offer`, `rejected`,
+...). `jobforge jobs` lists everything; a job with no reply after 30 days shows
+as **ghosted**.
+
+## Bring in past applications
+
+Already applied to jobs before Jobforge? Put them in a CSV (your agent can
+make one from your tracker) and import them, so `jobforge jobs` lists them and
+`prepare` warns you before applying to the same posting twice.
+
+```
+company,title,location,url,applied,status,status_date,notes
+Prairie Freight Co.,Operations Supervisor,"Springfield, IL",https://example.com/jobs/1,2026-09-02,rejected,2026-09-20,
+Cedar Mill Foods,Shift Supervisor,"Decatur, IL",,2026-09-10,applied,,
+```
+
+`company`, `title`, and `applied` are required. Write dates as YYYY-MM-DD. A
+status after `applied` (`screen`, `interview`, `rejected`, ...) needs the date
+it happened in `status_date`; if you do not know it, import the row as
+`applied` and record the rest later with `jobforge applied`.
+
+1. `jobforge import past.csv` shows what each row will become and writes
+   nothing. Check it.
+2. `jobforge import past.csv --write` creates the job folders. Running it again
+   skips rows already imported.
+
+Imported jobs hold only what happened: no posting text, plan, or documents.
+
+## Commands
+
+| Command | Does |
+|---|---|
+| `jobforge init <folder>` | Make a workspace (`--refresh` regenerates the agent files) |
+| `jobforge confirm` | Review drafted facts: keep, drop, or skip |
+| `jobforge inventory` | List your confirmed facts |
+| `jobforge prepare --from-url <url>` | Save a posting (or `--from-text <file>` with `--company` and `--title`) |
+| `jobforge check <job>` | Check a plan without writing documents |
+| `jobforge apply <job>` | Check, then write the resume and cover letter |
+| `jobforge applied --mark <job> --status <stage>` | Record what happened |
+| `jobforge jobs` | List your jobs and their status (`search` works too) |
+| `jobforge import <file.csv>` | Preview past applications from a CSV; `--write` imports them |
+
+`<job>` is the job folder's name or any unique piece of it.
+
+## Optional extras
+
+Not needed to use Jobforge. Planned for later releases:
+
+- **PDF output** (via Tectonic). Upload the `.docx` meanwhile; it is the
+  format applicant tracking systems read most reliably.
+- **A progress page** showing applications over time, callback rate, and
+  outcomes.
+
+## For maintainers
+
+```
+pip install -e ".[dev]"
+python -m pytest                 # also works uninstalled: pytest finds src/
+python -m coverage run -m pytest && python -m coverage report
+```
+
+- `docs/COMPONENTS.md`: what each part is called and where it lives
+- `docs/MVP.md`: what the first public release does, and build status
+- `docs/DECISIONS.md`: every scope decision; the governing record
+- `docs/SOW.md`: phases to 1.0
+- `docs/MAINTAINING.md`: how work is assigned and merged
+- `CONTRIBUTING.md`: issues welcome; outside PRs closed until the regression
+  harness exists (D-19)
+- `SECURITY.md`: how to report a problem that could expose a user's data
+- `docs/VM_CHECKLIST.md`: the clean-machine check before every release

jobforge-0.1.0rc2/README.md

@@ -0,0 +1,161 @@
+# Jobforge
+
+[![tests](https://github.com/jesse-anderson/jobforge/actions/workflows/tests.yml/badge.svg)](https://github.com/jesse-anderson/jobforge/actions/workflows/tests.yml)
+[![coverage](https://img.shields.io/endpoint?url=https://jesse-anderson.github.io/jobforge/coverage-badge.json)](https://github.com/jesse-anderson/jobforge/actions/workflows/tests.yml)
+[![python](https://img.shields.io/badge/python-3.11%2B-blue)](https://www.python.org/downloads/)
+[![license](https://img.shields.io/badge/license-PolyForm%20Noncommercial%201.0.0-blue)](https://github.com/jesse-anderson/jobforge/blob/main/LICENSE)
+[![platforms](https://img.shields.io/badge/platforms-Windows%20%7C%20Linux-lightgrey)](https://github.com/jesse-anderson/jobforge#what-you-need)
+[![dependencies](https://img.shields.io/badge/runtime%20dependencies-none-brightgreen)](https://github.com/jesse-anderson/jobforge/blob/main/pyproject.toml)
+
+Turn your real work history into a resume and cover letter tailored to one
+job posting, on your own computer, for free, without the tool ever stating a
+fact you did not give it.
+
+**Status: pre-release.** It works end to end, but it is not on PyPI yet.
+
+- **Free and local.** Nothing runs on a server. Your AI coding agent (Claude
+  Code first; Codex, Gemini CLI and others also work) does the writing, and
+  Jobforge does everything else.
+- **No invented facts.** Every number, employer, tool, and credential in your
+  documents is checked against facts you confirmed. If your agent invents
+  one, Jobforge refuses to write the document.
+- **Built for applicant tracking systems.** It writes `.docx` files with no
+  tables, columns, or hidden characters, then reads them back to prove what a
+  parser will see.
+
+Source-available under [PolyForm Noncommercial 1.0.0](https://github.com/jesse-anderson/jobforge/blob/main/LICENSE): free for personal use
+and for nonprofits, schools, and public institutions. Not for commercial use.
+
+## What you need
+
+- Windows or Linux (macOS is untested).
+- Python 3.11 or newer. Check with `python3 --version` on Linux, or
+  `python --version` on Windows.
+- An AI coding agent that runs in a terminal, such as Claude Code.
+
+## Set up (once)
+
+On Linux, install with pipx. Ubuntu, Debian, and others block `pip` for the
+system Python, and pipx is how they want apps like this one installed:
+
+```
+sudo apt install pipx
+pipx ensurepath                 # then open a new terminal
+pipx install jobforge           # or: git clone, then `pipx install .`
+jobforge init ~/jobsearch
+```
+
+On Windows:
+
+```
+pip install jobforge            # or: git clone, then `pip install .`
+jobforge init ~/jobsearch
+```
+
+That makes a folder with everything Jobforge needs:
+
+```
+jobsearch/
+  CLAUDE.md, AGENTS.md   instructions for your agent (generated)
+  profile/profile.toml   your name and contact details
+  profile/atoms/         your confirmed facts
+  profile/pending/       facts your agent drafted, waiting for you
+  intake/                put your old resume here
+  jobs/                  one folder per job you apply to
+```
+
+1. Copy your old resume (and any cover letters) into `intake/`.
+2. Open your agent in the `jobsearch` folder and say **"set me up"**. It reads
+   your resume, fills in your profile, and drafts your facts into
+   `profile/pending/`. Where your resume is unclear, it asks.
+3. Run **`jobforge confirm`**. For each drafted fact, you choose keep, drop,
+   or skip. Only facts you keep are ever used.
+
+To remove Jobforge later, `pipx uninstall jobforge` (Linux) or
+`pip uninstall jobforge` (Windows). Your `jobsearch` folder stays as it is.
+
+## Apply to a job
+
+Tell your agent: **"apply to this job: <link>"** (or paste the posting). It
+will:
+
+1. `jobforge prepare --from-url <link>` to save the posting,
+2. write a plan for which facts to use and how to word them,
+3. `jobforge check <job>` until the plan passes,
+4. `jobforge apply <job>`, which writes `Your_Name_Resume.docx` and
+   `Your_Name_Cover_Letter.docx` into the job's folder.
+
+Read both documents, then upload the `.docx` files yourself. `cover_letter.md`
+is a working copy for review; never send it.
+
+When you have applied: `jobforge applied --mark <job> --status applied`.
+Record later news the same way (`screen`, `interview`, `offer`, `rejected`,
+...). `jobforge jobs` lists everything; a job with no reply after 30 days shows
+as **ghosted**.
+
+## Bring in past applications
+
+Already applied to jobs before Jobforge? Put them in a CSV (your agent can
+make one from your tracker) and import them, so `jobforge jobs` lists them and
+`prepare` warns you before applying to the same posting twice.
+
+```
+company,title,location,url,applied,status,status_date,notes
+Prairie Freight Co.,Operations Supervisor,"Springfield, IL",https://example.com/jobs/1,2026-09-02,rejected,2026-09-20,
+Cedar Mill Foods,Shift Supervisor,"Decatur, IL",,2026-09-10,applied,,
+```
+
+`company`, `title`, and `applied` are required. Write dates as YYYY-MM-DD. A
+status after `applied` (`screen`, `interview`, `rejected`, ...) needs the date
+it happened in `status_date`; if you do not know it, import the row as
+`applied` and record the rest later with `jobforge applied`.
+
+1. `jobforge import past.csv` shows what each row will become and writes
+   nothing. Check it.
+2. `jobforge import past.csv --write` creates the job folders. Running it again
+   skips rows already imported.
+
+Imported jobs hold only what happened: no posting text, plan, or documents.
+
+## Commands
+
+| Command | Does |
+|---|---|
+| `jobforge init <folder>` | Make a workspace (`--refresh` regenerates the agent files) |
+| `jobforge confirm` | Review drafted facts: keep, drop, or skip |
+| `jobforge inventory` | List your confirmed facts |
+| `jobforge prepare --from-url <url>` | Save a posting (or `--from-text <file>` with `--company` and `--title`) |
+| `jobforge check <job>` | Check a plan without writing documents |
+| `jobforge apply <job>` | Check, then write the resume and cover letter |
+| `jobforge applied --mark <job> --status <stage>` | Record what happened |
+| `jobforge jobs` | List your jobs and their status (`search` works too) |
+| `jobforge import <file.csv>` | Preview past applications from a CSV; `--write` imports them |
+
+`<job>` is the job folder's name or any unique piece of it.
+
+## Optional extras
+
+Not needed to use Jobforge. Planned for later releases:
+
+- **PDF output** (via Tectonic). Upload the `.docx` meanwhile; it is the
+  format applicant tracking systems read most reliably.
+- **A progress page** showing applications over time, callback rate, and
+  outcomes.
+
+## For maintainers
+
+```
+pip install -e ".[dev]"
+python -m pytest                 # also works uninstalled: pytest finds src/
+python -m coverage run -m pytest && python -m coverage report
+```
+
+- `docs/COMPONENTS.md`: what each part is called and where it lives
+- `docs/MVP.md`: what the first public release does, and build status
+- `docs/DECISIONS.md`: every scope decision; the governing record
+- `docs/SOW.md`: phases to 1.0
+- `docs/MAINTAINING.md`: how work is assigned and merged
+- `CONTRIBUTING.md`: issues welcome; outside PRs closed until the regression
+  harness exists (D-19)
+- `SECURITY.md`: how to report a problem that could expose a user's data
+- `docs/VM_CHECKLIST.md`: the clean-machine check before every release

jobforge-0.1.0rc2/pyproject.toml

@@ -0,0 +1,70 @@
+[build-system]
+requires = ["setuptools>=77"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "jobforge"
+dynamic = ["version"]
+description = "Tailored resumes and cover letters from your real work history, run locally, with every fact checked."
+readme = "README.md"
+requires-python = ">=3.11"
+license = "LicenseRef-PolyForm-Noncommercial-1.0.0"
+license-files = ["LICENSE"]
+authors = [{ name = "Jesse Anderson" }, { name = "Mike Wilson" }]
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "Operating System :: Microsoft :: Windows",
+    "Operating System :: POSIX :: Linux",
+    "Environment :: Console",
+]
+# The runtime is stdlib only (D-13). Nothing goes in `dependencies`.
+dependencies = []
+
+[project.urls]
+Homepage = "https://github.com/jesse-anderson/jobforge"
+Source = "https://github.com/jesse-anderson/jobforge"
+Issues = "https://github.com/jesse-anderson/jobforge/issues"
+Security = "https://github.com/jesse-anderson/jobforge/security/policy"
+
+[project.optional-dependencies]
+dev = ["pytest>=8", "coverage>=7"]
+
+[project.scripts]
+jobforge = "jobforge.cli:main"
+
+[tool.setuptools.dynamic]
+version = { attr = "jobforge.__version__" }
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+jobforge = ["agent/*.md"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+pythonpath = ["src"]
+addopts = "-ra"
+markers = [
+    "critical: D-30 critical path; must stay at 100% coverage",
+]
+
+[tool.coverage.run]
+source = ["jobforge"]
+branch = true
+
+[tool.coverage.report]
+show_missing = true
+
+# Critical paths (D-30, D-51): held to 100% line and branch coverage by
+# `python tools/critical_coverage.py`, in CI and before every push.
+[tool.jobforge]
+critical = [
+    "workspace.py", "profile.py", "agent/__init__.py",   # setup
+    "extract.py", "prepare.py",                          # job intake
+    "plan.py",                                           # plan validation
+    "render.py", "docx.py",                              # rendering
+    "scan.py", "factcheck.py", "facts.py", "apply.py",   # safety
+    "ledger.py", "importer.py",                          # ledger
+    "confirm.py",                                        # facts going live
+]

jobforge-0.1.0rc2/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

jobforge-0.1.0rc2/src/jobforge/__init__.py

@@ -0,0 +1,3 @@
+"""Jobforge: tailored resumes and cover letters from facts you confirmed."""
+
+__version__ = "0.1.0rc2"

jobforge-0.1.0rc2/src/jobforge/__main__.py

@@ -0,0 +1,5 @@
+import sys
+
+from .cli import main
+
+sys.exit(main())

jobforge-0.1.0rc2/src/jobforge/agent/__init__.py

@@ -0,0 +1,117 @@
+"""The agent layer: generated instruction files for the user's agent.
+
+One canonical source, `instructions.md`, renders to every agent file, so
+they cannot drift apart (D-11). Limits in the text come from the user's
+profile, so the instructions and the validator always state the same number
+(FINDINGS F19). `jobforge init` writes these into the workspace, since a user
+who installed from PyPI never sees the repo (SOW A-18), and `init --refresh`
+rewrites them after an upgrade or a profile change.
+"""
+from __future__ import annotations
+
+import hashlib
+import re
+from importlib import resources
+
+from ..errors import JobforgeError
+
+AGENT_FILES = ("CLAUDE.md", "AGENTS.md")
+GENERATED_HEADER = ("<!-- Generated by Jobforge from its built-in instructions. Edit only "
+                    "below the\n     \"Your notes\" line at the end; `jobforge init "
+                    "--refresh` regenerates the rest. -->\n\n")
+_PLACEHOLDER = re.compile(r"\{\{(\w+)\}\}")
+
+# The user's own section (#33). Everything below this line survives a
+# refresh. The hash fingerprints the generated part above it, so a refresh
+# can tell whether someone edited that part, and back it up if so.
+NOTES_MARKER = "Your notes: write below this line"
+_NOTES_LINE = re.compile(rf"^<!-- {re.escape(NOTES_MARKER)}\..*?\(jobforge:([0-9a-f]+)\) -->\n?",
+                         re.MULTILINE)
+_LEADING_COMMENT = re.compile(r"\A\s*<!--.*?-->\s*", re.DOTALL)
+
+
+def _fingerprint(generated: str) -> str:
+    return hashlib.sha256(generated.rstrip().encode("utf-8")).hexdigest()[:12]
+
+
+def _with_notes(generated: str, notes: str = "") -> str:
+    return (f"{generated}\n<!-- {NOTES_MARKER}. `jobforge init --refresh` keeps everything "
+            f"below it. (jobforge:{_fingerprint(generated)}) -->\n{notes}")
+
+
+def source_text() -> str:
+    return resources.files(__package__).joinpath("instructions.md").read_text(encoding="utf-8")
+
+
+def _generated(values: dict[str, object]) -> str:
+    def sub(m: re.Match) -> str:
+        key = m.group(1)
+        if key not in values:
+            raise KeyError(f"instructions.md uses {{{{{key}}}}} but no value was given")
+        return str(values[key])
+    return GENERATED_HEADER + _PLACEHOLDER.sub(sub, source_text())
+
+
+def render(values: dict[str, object]) -> str:
+    """A fresh agent file: the generated text, then an empty notes section."""
+    return _with_notes(_generated(values))
+
+
+def values_for(ws) -> dict[str, object]:
+    """Values from the workspace profile, or defaults while the profile is
+    still the empty starter."""
+    from ..profile import Profile, load_profile
+    try:
+        profile = load_profile(ws)
+    except JobforgeError:
+        profile = Profile(name="")
+    return {
+        "name": profile.name or "the user",
+        "cover_letter_max_words": profile.cover_letter_max_words,
+        "page_limit": profile.page_limit,
+        "ghost_after_days": profile.ghost_after_days,
+        "dash_rule": ", with no em or en dashes" if profile.ban_dashes else "",
+    }
+
+
+def _split(old: str, body: str) -> tuple[str, bool]:
+    """An existing agent file -> (the user's notes, whether the generated
+    part was edited). A file from before the notes section (#33) counts as
+    edited unless it is this version's text."""
+    m = _NOTES_LINE.search(old)
+    if m:
+        return old[m.end():], _fingerprint(old[:m.start()]) != m.group(1)
+    return "", _LEADING_COMMENT.sub("", old, count=1).rstrip() != body.rstrip()
+
+
+def _free_backup(path):
+    n = 1
+    while True:
+        bak = path.with_name(path.name + (".bak" if n == 1 else f".bak{n}"))
+        if not bak.exists():
+            return bak
+        n += 1
+
+
+def write_agent_files(ws) -> list[str]:
+    """Write CLAUDE.md and AGENTS.md, keeping the user's notes section. An
+    edit above it is never dropped silently: the old file is saved as
+    <name>.bak (then .bak2, ...) first. Returns a notice per backup."""
+    generated = _generated(values_for(ws))
+    body = _LEADING_COMMENT.sub("", generated, count=1)
+    notices = []
+    for name in AGENT_FILES:
+        path = ws.root / name
+        notes = ""
+        if path.exists():
+            old = path.read_text(encoding="utf-8-sig")
+            notes, edited = _split(old, body)
+            if edited:
+                bak = _free_backup(path)
+                bak.write_text(old, encoding="utf-8", newline="\n")
+                notices.append(
+                    f"{name} had edits outside its notes section, so the old file is "
+                    f"saved as {bak.name}. Move anything you want to keep below the "
+                    f"\"Your notes\" line at the end of {name}.")
+        path.write_text(_with_notes(generated, notes), encoding="utf-8", newline="\n")
+    return notices