Autoresearch — Autonomous Goal-directed Iteration
Safety Invariants (all subcommands)
- Never push, publish, or deploy without explicit user approval.
- Bounded by default. Override with
Iterations: unlimited.
- All results logged to
autoresearch/{subcommand}-{YYMMDD}-{HHMM}/ directory.
- Chain handoff via
handoff.json. Evals reads *-results.tsv.
Dispatch (bare /autoresearch)
Parse the invocation in this order:
| Condition |
Mode |
Metric: or Verify: present |
Classic — existing metric loop, unchanged |
| Free-form natural-language goal, no metric/verify |
Orchestrator — see Orchestrator section |
| Nothing |
Setup wizard — interactive config builder |
--classic flag |
Force Classic regardless of goal text |
--auto flag |
Force Orchestrator regardless of goal text |
Print a banner on every invocation: [autoresearch] mode: classic | orchestrator | wizard.
Subcommands
| Command |
Does |
Default Iterations |
/autoresearch |
Iterate against a metric: modify → verify → keep/discard |
25 |
/autoresearch_plan |
Convert a goal into validated Scope, Metric, Verify config |
N/A |
/autoresearch_debug |
Hunt bugs: hypothesize → test → falsify → repeat |
15 |
/autoresearch_fix |
Crush errors one-by-one until zero remain |
20 |
/autoresearch_security |
STRIDE + OWASP audit with red-team personas |
15 |
/autoresearch_ship |
Ship through 8 phases: checklist → dry-run → deploy → verify |
N/A |
/autoresearch_scenario |
Generate edge cases across 12 dimensions |
20 |
/autoresearch_predict |
5 expert personas debate before implementation |
N/A |
/autoresearch_learn |
Scout codebase → generate docs or wiki → validate → fix loop |
10 |
/autoresearch_reason |
Adversarial debate with blind judges until convergence |
8 |
/autoresearch_probe |
8 personas interrogate requirements until saturation |
15 |
/autoresearch_improve |
Research ICP challenges, discover improvements, generate PRDs |
15 |
/autoresearch_evals |
Analyze iteration results: trends, plateaus, regressions |
N/A |
/autoresearch_regression |
Regression stability gate: baseline vs candidate, verdict STABLE/UNSTABLE |
N/A |
Universal Flags
| Flag |
Applies To |
Purpose |
Iterations: N |
All looping |
Set iteration count |
Iterations: unlimited |
All looping |
Opt-in unbounded |
--evals |
All looping |
Mid-loop checkpoints + final summary |
--evals-interval N |
All looping |
Override checkpoint frequency |
--chain <targets> |
All |
Sequential handoff after completion |
--<subcommand> |
All |
Shorthand for --chain <subcommand> |
--dry-run |
Orchestrator |
Print derived config + planned pipeline; no execution |
--max-cycles N |
Orchestrator |
Hard ceiling on orchestration cycles (default 50) |
--classic |
Bare /autoresearch |
Force Classic metric-loop mode |
--auto |
Bare /autoresearch |
Force Orchestrator mode |
Orchestrator
Activated when a plain-language goal is given without Metric:/Verify:. Classifies the goal into a Goal archetype — see references/orchestrator-routing.md for the archetype table and router decision table.
Resolve every scripts/... path below relative to this installed skill directory, never relative to the caller's working directory.
Two modes based on archetype:
- Orchestration loop — predicate-bearing archetypes (ship-ready, optimize-metric, fix-broken, harden, build-feature, explore). Goal has a mechanical Success predicate; the loop runs until that predicate is met.
- Single-pass dispatch — subjective/terminal archetypes (document, what-to-build, decide-design). Routes once to the fitting subcommand (learn / improve / reason), lets it self-terminate, then reports. No loop, no Plateau, no ship gate.
Orchestration Loop Steps
Backed by scripts/orchestrate.sh (deterministic seam — all routing logic lives there). Subcommands exposed: classify, next-hop, units, plateau, screen-cmd, verdict, validate-state, screen-state-predicate.
- Classify —
scripts/orchestrate.sh classify "<goal>" → archetype label + mode.
- Derive predicate — reuse
plan logic to produce a concrete Success predicate: exact shell command + expected output. For optimize-metric, run the full plan/wizard derivation internally.
- Confirm — ONE
question showing: archetype, mode, concrete predicate (command + expected output), terminal choice (stop-at-verified vs proceed-to-ship). Misclassifications are caught here, not mid-run.
- Round-0 dry-run — prove the predicate command runs and returns a value; safety-screen every derived command via
screen-cmd; print projected cycle budget. Stop here if --dry-run.
- Loop until predicate satisfied:
a. Assess state via cheap signals (last
handoff.json, regression verdict, error count) + affected-test verify.
b. scripts/orchestrate.sh next-hop orchestrator-state.json → next subcommand.
c. Run subcommand (its own bounded inner loop).
d. Record per-hop outcome ∈ {progressed, no-op, failed, blocked}.
e. Fold hop's handoff.json into orchestrator-state.json.
f. scripts/orchestrate.sh units → recompute Units remaining.
- Stop conditions (checked after each hop):
- Predicate met → ship gate (only if ship is in the pipeline) else
CONVERGED.
scripts/orchestrate.sh plateau orchestrator-state.json → true → stop + report PLATEAU.- Cycles > ceiling (default 50, override
--max-cycles N) → stop + report CEILING.
- Hop outcome
blocked/failed with no alternative route → checkpoint + stop + report BLOCKED.
Orchestrator State
orchestrator-state.json — orchestrator-owned, additive. Tracks: goal, archetype, predicate, terminal-choice, units_remaining history, cycle count, per-hop pipeline log with outcomes, current incumbent. Each hop's handoff.json is unchanged (single-hop bridge); the orchestrator reads it and folds it in. Two clearly-owned state objects, no overlap.
Orchestrator Safety Invariants
- Never auto-approve ship/deploy/push. The orchestrator never passes
--auto to ship; deploy always requires explicit user approval.
- Data-migration behind anchored DB-URL allowlist. Reuses regression's allowlist — host must be
localhost/127.0.0.1/container hostname, or database name carries _test/_ci suffix. Bare substring match does not qualify. Anything else refused.
- screen-cmd on every derived command — run before the loop starts AND on every command read from a persisted state file on resume. Persisted commands are never trusted; resume re-screens the pinned predicate via
screen-state-predicate and refuses on refuse.
- No un-screened commands mid-loop. The autonomous loop cannot introduce new shell commands that bypass
screen-cmd.
- Predicate pinned, not re-derived. Round-0 writes the derived Success predicate verbatim into
orchestrator-state.json; every cycle and every resume reuses that exact string so "done" is reproducible across runs.
- Validate the ledger before routing.
validate-state gates orchestrator-state.json (required fields + coarse types); a malformed ledger is not trusted to route from.
- Independent verify before convergence. High-impact changes accepted on the working signal set
pending_verify; next-hop routes to a verify hop (held-out / adversarial check) before DONE or ship. The verify hop never auto-approves ship.
- Unknown-units cycles excluded from Plateau counter. A cycle where
units returns unknown (e.g. runner crash) is not counted as zero-progress; repeated unknown routes to BLOCKED.
1---2name: autoresearch-63description: Autonomous iteration loop: modify, verify, keep/discard against any metric4---56# Autoresearch — Autonomous Goal-directed Iteration78## Safety Invariants (all subcommands)9- Never push, publish, or deploy without explicit user approval.10- Bounded by default. Override with `Iterations: unlimited`.11- All results logged to `autoresearch/{subcommand}-{YYMMDD}-{HHMM}/` directory.12- Chain handoff via `handoff.json`. Evals reads `*-results.tsv`.1314## Dispatch (bare `/autoresearch`)1516Parse the invocation in this order:1718| Condition | Mode |19|---|---|20| `Metric:` or `Verify:` present | **Classic** — existing metric loop, unchanged |21| Free-form natural-language goal, no metric/verify | **Orchestrator** — see Orchestrator section |22| Nothing | **Setup wizard** — interactive config builder |23| `--classic` flag | Force Classic regardless of goal text |24| `--auto` flag | Force Orchestrator regardless of goal text |2526Print a banner on every invocation: `[autoresearch] mode: classic | orchestrator | wizard`.2728## Subcommands2930| Command | Does | Default Iterations |31|---|---|---|32| `/autoresearch` | Iterate against a metric: modify → verify → keep/discard | 25 |33| `/autoresearch_plan` | Convert a goal into validated Scope, Metric, Verify config | N/A |34| `/autoresearch_debug` | Hunt bugs: hypothesize → test → falsify → repeat | 15 |35| `/autoresearch_fix` | Crush errors one-by-one until zero remain | 20 |36| `/autoresearch_security` | STRIDE + OWASP audit with red-team personas | 15 |37| `/autoresearch_ship` | Ship through 8 phases: checklist → dry-run → deploy → verify | N/A |38| `/autoresearch_scenario` | Generate edge cases across 12 dimensions | 20 |39| `/autoresearch_predict` | 5 expert personas debate before implementation | N/A |40| `/autoresearch_learn` | Scout codebase → generate docs or wiki → validate → fix loop | 10 |41| `/autoresearch_reason` | Adversarial debate with blind judges until convergence | 8 |42| `/autoresearch_probe` | 8 personas interrogate requirements until saturation | 15 |43| `/autoresearch_improve` | Research ICP challenges, discover improvements, generate PRDs | 15 |44| `/autoresearch_evals` | Analyze iteration results: trends, plateaus, regressions | N/A |45| `/autoresearch_regression` | Regression stability gate: baseline vs candidate, verdict STABLE/UNSTABLE | N/A |4647## Universal Flags4849| Flag | Applies To | Purpose |50|---|---|---|51| `Iterations: N` | All looping | Set iteration count |52| `Iterations: unlimited` | All looping | Opt-in unbounded |53| `--evals` | All looping | Mid-loop checkpoints + final summary |54| `--evals-interval N` | All looping | Override checkpoint frequency |55| `--chain <targets>` | All | Sequential handoff after completion |56| `--<subcommand>` | All | Shorthand for `--chain <subcommand>` |57| `--dry-run` | Orchestrator | Print derived config + planned pipeline; no execution |58| `--max-cycles N` | Orchestrator | Hard ceiling on orchestration cycles (default 50) |59| `--classic` | Bare `/autoresearch` | Force Classic metric-loop mode |60| `--auto` | Bare `/autoresearch` | Force Orchestrator mode |6162## Orchestrator6364Activated when a plain-language goal is given without `Metric:`/`Verify:`. Classifies the goal into a **Goal archetype** — see `references/orchestrator-routing.md` for the archetype table and router decision table.6566Resolve every `scripts/...` path below relative to this installed skill directory, never relative to the caller's working directory.6768**Two modes based on archetype:**69- **Orchestration loop** — predicate-bearing archetypes (ship-ready, optimize-metric, fix-broken, harden, build-feature, explore). Goal has a mechanical Success predicate; the loop runs until that predicate is met.70- **Single-pass dispatch** — subjective/terminal archetypes (document, what-to-build, decide-design). Routes once to the fitting subcommand (learn / improve / reason), lets it self-terminate, then reports. No loop, no Plateau, no ship gate.7172### Orchestration Loop Steps7374Backed by `scripts/orchestrate.sh` (deterministic seam — all routing logic lives there). Subcommands exposed: `classify`, `next-hop`, `units`, `plateau`, `screen-cmd`, `verdict`, `validate-state`, `screen-state-predicate`.75761. **Classify** — `scripts/orchestrate.sh classify "<goal>"` → archetype label + mode.772. **Derive predicate** — reuse `plan` logic to produce a concrete Success predicate: exact shell command + expected output. For `optimize-metric`, run the full plan/wizard derivation internally.783. **Confirm** — ONE `question` showing: archetype, mode, concrete predicate (command + expected output), terminal choice (stop-at-verified vs proceed-to-ship). Misclassifications are caught here, not mid-run.794. **Round-0 dry-run** — prove the predicate command runs and returns a value; safety-screen every derived command via `screen-cmd`; print projected cycle budget. Stop here if `--dry-run`.805. **Loop** until predicate satisfied:81 a. Assess state via cheap signals (last `handoff.json`, regression verdict, error count) + affected-test verify.82 b. `scripts/orchestrate.sh next-hop orchestrator-state.json` → next subcommand.83 c. Run subcommand (its own bounded inner loop).84 d. Record per-hop outcome ∈ {progressed, no-op, failed, blocked}.85 e. Fold hop's `handoff.json` into `orchestrator-state.json`.86 f. `scripts/orchestrate.sh units` → recompute **Units remaining**.876. **Stop conditions** (checked after each hop):88 - Predicate met → ship gate (only if ship is in the pipeline) else `CONVERGED`.89 - `scripts/orchestrate.sh plateau orchestrator-state.json` → true → stop + report `PLATEAU`.90 - Cycles > ceiling (default 50, override `--max-cycles N`) → stop + report `CEILING`.91 - Hop outcome `blocked`/`failed` with no alternative route → checkpoint + stop + report `BLOCKED`.9293### Orchestrator State9495`orchestrator-state.json` — orchestrator-owned, additive. Tracks: goal, archetype, predicate, terminal-choice, `units_remaining` history, cycle count, per-hop pipeline log with outcomes, current incumbent. Each hop's `handoff.json` is unchanged (single-hop bridge); the orchestrator reads it and folds it in. Two clearly-owned state objects, no overlap.9697### Orchestrator Safety Invariants9899- **Never auto-approve ship/deploy/push.** The orchestrator never passes `--auto` to `ship`; deploy always requires explicit user approval.100- **Data-migration behind anchored DB-URL allowlist.** Reuses regression's allowlist — host must be `localhost`/`127.0.0.1`/container hostname, or database name carries `_test`/`_ci` suffix. Bare substring match does not qualify. Anything else refused.101- **screen-cmd on every derived command** — run before the loop starts AND on every command read from a persisted state file on resume. Persisted commands are never trusted; resume re-screens the pinned predicate via `screen-state-predicate` and refuses on `refuse`.102- **No un-screened commands mid-loop.** The autonomous loop cannot introduce new shell commands that bypass `screen-cmd`.103- **Predicate pinned, not re-derived.** Round-0 writes the derived Success predicate verbatim into `orchestrator-state.json`; every cycle and every resume reuses that exact string so "done" is reproducible across runs.104- **Validate the ledger before routing.** `validate-state` gates `orchestrator-state.json` (required fields + coarse types); a malformed ledger is not trusted to route from.105- **Independent verify before convergence.** High-impact changes accepted on the working signal set `pending_verify`; `next-hop` routes to a `verify` hop (held-out / adversarial check) before `DONE` or ship. The verify hop never auto-approves ship.106- **Unknown-units cycles excluded from Plateau counter.** A cycle where `units` returns `unknown` (e.g. runner crash) is not counted as zero-progress; repeated `unknown` routes to `BLOCKED`.
gabrielmoreira