adopt

Run the SpecSelf adopt pass: read a body of notes the human already has and draft their first corpus as candidates, so someone arriving with years of notes does not have to author a foundation from scratch. Synthesize in the Journalist lens (personas/journalist.md). Honor AGENTS.md.

Read first: AGENTS.md (the one rule and the hard boundaries), then steering/ (methodology, privacy, tone), then the templates/ you will write into (principle.md, aspirational.md, area.md, goal.md, project.md, habit.md).

Adopt does not interview. It reads what already exists and proposes candidates. The stories behind a value rarely live in notes, so adopt is a head start, not a replacement for the foundation sitting. Say so in the closing report.

The contract (read this before touching anything)

• The human names a source directory. Ask for it, and ask for any paths inside it that are off-limits.
• The source directory is READ-ONLY, always. List, stat, and read its files. Never write, edit, move, rename, or delete anything inside it. It is not yours to change.
• You write only into your-life/. Every candidate artifact lands under your-life/, through the templates, as a <!-- CANDIDATE - edit or delete --> block. That is the one place adopt writes.
• Never read your-life/.sensitive/ (a standing boundary), and never read any source path the human marked off-limits.
• Never copy the notes wholesale. Adopt drafts a bounded set of candidates with citations. It does not import a vault into your-life/.

Steps

1. Get the source and the off-limits list. Ask the human for the source directory and for any files or subfolders within it that are private and must not be read. Record the off-limits list; you will name it in the report.
2. Bound the sample. Do not read the whole directory. Sample the 50 most recently modified files plus any specific files the human names as worth reading. This keeps a giant vault from blowing the session. Skip anything under an off-limits path and skip your-life/.sensitive/ entirely. If a file cannot be read (binary, encoding, permissions), skip it and note it for the report.
3. Read read-only and gather evidence. Read the sampled files. For each theme, role, outcome, or routine that recurs, note where it appears (the file path) and any date you can derive from the note content (a dated entry, a heading date, a logged event). A date written in the note is what lets a principle be revealed; a file's modified timestamp is provenance for the citation, not evidence of when something happened.

4. Draft candidates through the templates. Write each candidate into your-life/ as a <!-- CANDIDATE - edit or delete --> block, hydrating the matching template, and cite its source in the block (the file path, plus a date where one is derivable from the content). Draft, in bounds:

   Candidate	Template	Where it comes from	Bound
   Revealed principles	templates/principle.md	recurring stances the notes show dated evidence for	3-5
   Aspirational entries	templates/aspirational.md	stances the notes state but do not back with a dated instance	0-3
   Areas	templates/area.md	standing roles the notes maintain (health, money, a craft, a relationship)	2-5
   Goals	templates/goal.md	finite outcomes the notes are working toward, each with a success criterion and a serves: link	2-4
   Projects	templates/project.md	concrete efforts the notes show in flight now	at most 3
   Habits	templates/habit.md	routines the notes track or repeat	at most 3

   • The bounds are ceilings, not quotas. Fewer is fine. Drafting three thin principles to hit the floor flatters the human with volume and violates the AGENTS.md boundary against fabricated evidence. Draft only what the notes actually support.
   • revealed requires a dated instance from the notes. A principle candidate is kind: revealed only when you can cite at least one concrete, dated instance from the sampled notes under its ## Evidence section. Without that, it is not revealed. Propose it for aspirational.md instead, with its promotion bar, or skip it. Never invent a date to clear the bar, and never collapse the gap between who the notes show the human to be and who they want to be.
   • Do not draft a vision. The five-year portrait is exactly what notes do not contain; leave it for the foundation sitting and name it in the report.
5. Write the adopt report. Close with a short report (in chat, not a corpus file) covering:
   • Drawn from where. Which candidates came from which files, so the human can trace every draft back to its source.
   • Skipped and why. What was left out and the reason: out of the sample (older than the 50 most recent, not named), marked off-limits, or unreadable.
   • What the notes could not cover. Which foundation phases have no material here. The vision and the lived stories behind the load-bearing values almost never live in notes, so those phases stay empty until the human does them.
   • The recommendation. Sitting 1 of /foundation (Operating System + Load-Bearing Values) is still worth running, because it captures exactly the decision-making and value stories the notes lack. Say this plainly.

Placement

/adopt can run before /foundation (turn the notes into a starting draft, then let the sittings deepen it), after sitting 1 (keep the value stories from sitting 1, let adopt fill areas, goals, and projects from the notes), or instead of sittings 2-4 (adopt covers the horizons that live in notes; you still want sitting 1 for the stories). GETTING-STARTED carries this as its "Already have notes?" section.

Hard boundaries

• The source directory is read-only. No write, edit, move, rename, or delete inside it, ever. If a step seems to call for changing a source file, it is wrong; stop and re-read this section.
• your-life/.sensitive/ and off-limits paths are never read.
• No wholesale import. Adopt drafts bounded candidates with citations; it never copies a vault into your-life/.
• Never fabricate evidence and never manufacture a date to promote a candidate to revealed.

The never-authors rule

Write every proposed principle, area, goal, project, or habit as a <!-- CANDIDATE - edit or delete --> block, never committed text. Each candidate is something the human rewrites in their own words or discards; adopt surfaces and drafts from the notes, the human edits and commits. Nothing produced by this skill is authored as final on the human's behalf.