feat(20g): migrate all remaining agents to .profile + remove ops repo journal dirs

dev-bot commented

2026-04-01 06:32:43 +00:00

Collaborator

Parent

Part of #20 — agent .profile repo.

What to do

First: extract shared context setup into lib/formula-session.sh

The PR branch adds profile_load_lessons / profile_write_journal calls to every agent script, creating duplicate boilerplate. Instead, add a single shared function that each agent calls once:
```
# lib/formula-session.sh
formula_prepare_profile_context() {
  # Load lessons from .profile (if agent has one)
  profile_load_lessons
  LESSONS_INJECTION="${LESSONS_CONTEXT:-}"
}
```
Similarly, wrap the post-session journal into a single call each agent makes at the end. The goal: each agent script has ONE call to the shared function, not 3-4 lines of repeated variable assignments.

This avoids the duplicate-detection CI failure that blocked the previous attempt.
Run disinto hire-an-agent for each remaining agent:
- disinto hire-an-agent review-bot review
- disinto hire-an-agent planner-bot planner
- disinto hire-an-agent gardener-bot gardener
- disinto hire-an-agent supervisor-bot supervisor
- disinto hire-an-agent predictor-bot predictor
Wire each agent script to call the shared function (one line per script, not boilerplate)
Remove ops repo journal directories:
- Delete journal/planner/ and journal/supervisor/ from disinto-ops
- Update bin/disinto init to stop creating these dirs
Update AGENTS.md to document the .profile model

Why the previous PR failed

PR #106 introduced 14 new duplicate code blocks — the same LESSONS_INJECTION, SCRATCH_CONTEXT, profile_load_lessons setup repeated across planner-run.sh, gardener-run.sh, predictor-run.sh, supervisor-run.sh, and review scripts. The duplicate-detection CI step correctly flagged these.

The fix: extract the common pattern into a shared function in lib/formula-session.sh so each agent script calls one function instead of repeating the boilerplate.

Verification

Each agent has a .profile repo on Forgejo
Each agent reads formula from .profile at session start
Each agent calls formula_prepare_profile_context() (one line, no duplication)
Ops repo has no journal/ directories
AGENTS.md reflects the new model
duplicate-detection CI step passes (zero new duplicates)
CI green

Dependencies

Depends on #85 (formula loading), #87 (branch protection), #97 (generic journal aspect). #86 and #89 were absorbed into #97.

## Parent Part of #20 — agent .profile repo. ## What to do 1. **First: extract shared context setup into `lib/formula-session.sh`** The PR branch adds `profile_load_lessons` / `profile_write_journal` calls to every agent script, creating duplicate boilerplate. Instead, add a single shared function that each agent calls once: ```bash # lib/formula-session.sh formula_prepare_profile_context() { # Load lessons from .profile (if agent has one) profile_load_lessons LESSONS_INJECTION="${LESSONS_CONTEXT:-}" } ``` Similarly, wrap the post-session journal into a single call each agent makes at the end. The goal: each agent script has ONE call to the shared function, not 3-4 lines of repeated variable assignments. This avoids the duplicate-detection CI failure that blocked the previous attempt. 2. Run `disinto hire-an-agent` for each remaining agent: - `disinto hire-an-agent review-bot review` - `disinto hire-an-agent planner-bot planner` - `disinto hire-an-agent gardener-bot gardener` - `disinto hire-an-agent supervisor-bot supervisor` - `disinto hire-an-agent predictor-bot predictor` 3. Wire each agent script to call the shared function (one line per script, not boilerplate) 4. Remove ops repo journal directories: - Delete `journal/planner/` and `journal/supervisor/` from disinto-ops - Update `bin/disinto` init to stop creating these dirs 5. Update AGENTS.md to document the .profile model ## Why the previous PR failed PR #106 introduced 14 new duplicate code blocks — the same `LESSONS_INJECTION`, `SCRATCH_CONTEXT`, `profile_load_lessons` setup repeated across planner-run.sh, gardener-run.sh, predictor-run.sh, supervisor-run.sh, and review scripts. The duplicate-detection CI step correctly flagged these. The fix: extract the common pattern into a shared function in `lib/formula-session.sh` so each agent script calls one function instead of repeating the boilerplate. ## Verification - Each agent has a `.profile` repo on Forgejo - Each agent reads formula from `.profile` at session start - Each agent calls `formula_prepare_profile_context()` (one line, no duplication) - Ops repo has no `journal/` directories - AGENTS.md reflects the new model - **duplicate-detection CI step passes** (zero new duplicates) - CI green ## Dependencies Depends on #85 (formula loading), #87 (branch protection), #97 (generic journal aspect). #86 and #89 were absorbed into #97.