Backed by: state/lib/testimonials.ts + state/lib/email.ts (draft only)

testimonial-ask

Intersects commitment-audit silent-ships with clients whose pulse is green (no open complaints). Drafts a testimonial request. Same never-auto-send gate as dormant-ping.

Steps

1. Read latest commitment-audit output.
2. Filter: silent_shipped.length > 0 && client_pulse.status == "green".
3. Dispatch gemini to draft a specific-work request referencing the

shipped artifact.

1. voice.checkTone() gate on each draft.
2. Write to state/log/testimonial-ask/<date>/<slug>.md.

Eval

Same three-part shape as dormant-ping (shape_ok + never_sent + tone_pass). Score 1.0 / 0.5 / 0.0.

score("testimonial-ask", run_id, {
  score: shape_ok && never_sent && all_tone_ok ? 1.0 : ...,
  drafts_written: n,
  references_specific_work: drafts.filter(d => d.cites_artifact).length,
  primary_issue: ...,
});

Additional signal: references_specific_work must equal drafts_written. A generic "hey, would you write me a testimonial" draft fails because it doesn't cite the specific shipped thing.

Gotchas

• Never ask a yellow/red client. The pulse filter is the gate.
• Draft must reference the actual commit or artifact URL - that's why

it's chained off commitment-audit, not standalone.

Rubric

criteria:
  - name: drafts_written
    kind: deterministic
    check: "The number of markdown files written to 'state/log/testimonial-ask/<date>/<slug>.md' matches the count of eligible clients."
  - name: references_specific_work
    kind: judge
    check: "Each drafted testimonial request explicitly references a specific shipped artifact or commit URL."
  - name: no_yellow_red_clients
    kind: deterministic
    check: "No testimonial requests were drafted for clients whose 'client_pulse.status' was 'yellow' or 'red'."
  - name: never_auto_sent
    kind: deterministic
    check: "No email containing a testimonial request was automatically sent by the skill execution."

testimonial-ask - loader

Per-turn rules for the testimonial-ask skill. Full reference: state/skills/testimonial-ask/SKILL.md. Do not skip these.

Critical Rules

• NEVER ask a yellow/red client. The client_pulse green filter is the gate - non-negotiable.
• NEVER auto-send. Drafts only; same never-auto-send gate as dormant-ping.
• ALWAYS reference the actual commit / artifact URL - that's why this skill is chained off commitment-audit, never standalone.
• ALWAYS run voice.checkTone() on every draft before writing.
• A generic "hey, would you write me a testimonial" draft FAILS - references_specific_work must equal drafts_written.

Commands

| ui dashboard | state/skills/testimonial-ask/resources/ui.openui | |invoke: chain after commitment-audit → filter silent_shipped.length > 0 && client_pulse.status == "green" → dispatch gemini to draft → voice.checkTone() → write to state/log/testimonial-ask/<date>/<slug>.md |verify: re-read state/log/testimonial-ask/<date>/ and check tone-pass + cites-artifact on each |eval log: state/log/evals.ndjson (skill: "testimonial-ask") - required: drafts_written, references_specific_work

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/testimonial-ask/resources/ui.openui. Read it before rendering or editing this skill's generated component surface.
• Treat this resource as a first-class artifact of the skill, not a generic chat response. Improve it when the skill's user-facing output needs to become richer.
• System resources compose OpenUI primitives and inherit SnappyChat tokens. Use ui_contract: branded in SKILL.md only for deliberate platform or client visuals.

Known Pitfalls

• Score 1.0 requires shape_ok && never_sent && all_tone_ok && references_specific_work === drafts_written.
• Upstream candidate must already be HIGH-scored by testimonial-scan. LOW-scoring quotes are a hard reject - drafting against them burns the relationship for a useless asset.
• One quote per permission ask. Do NOT batch.
• Never re-ask a declined quote - that state lives in snappy-knowledge contact notes.

Self-Test

An agent reading this should correctly:

1. [ ] Refuse to draft against a yellow/red pulse client
2. [ ] Refuse to auto-send the draft
3. [ ] Cite the specific commit/artifact in every draft

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] testimonial-ask: <what was missing>" >> state/log/loader-feedback.log

<!-- kernel-ok: historical attribution for the PID-loader pattern. -->

Self-correcting loader (PID feedback)

This is the rule that makes loaders learn. Ported from snappy-kernel, where it demonstrably worked: the snappy-dom-cartographer skill mapped Skool admin UI once, every later ship-agent inherited the full map for free.

Before you finish, do two things:

1. Fix gaps (P - proportional). MANDATORY. If this loader didn't cover your case - if you had to read another file, run an undocumented command, work around a wrong selector, or discover a quirk - you MUST attempt an Edit to this AGENTS.md before you log.

• Default action: edit this file. Surgical. Add the missing line, table

row, or code example where it belongs. No restructuring.

• Only LOGGED is allowed when: the fix needs >10 lines, spans multiple

files, or requires a structural rewrite. In that case the state/regen/drain.sh queue picks it up asynchronously.

• "I didn't have time" / "it's minor" / "the next agent will figure it out"

are NOT valid reasons. Minor gaps compound - that's the whole point of this loop.

• Concrete example of a valid P-fix: discover agent-browser screenshot

takes path positionally not as --output <path>. One line in a table. Edit it in. Do not just log it.

• The goal: the next agent never has to leave the loader.

2. Log the result.

echo "[$(date -u +%FT%TZ)] testimonial-ask: <what was missing or fixed> [FIXED|LOGGED]" >> state/log/loader-feedback.log

• FIXED = you patched this loader inline (P-fix).
• LOGGED = too large for inline; the PostToolUse enqueue + Stop-hook drain

will rewrite the loader from scratch on next session-end.

Do not skip this. Every agent run must leave the system better than it found it. The loader is the setpoint; you are the sensor; the gap is the error signal; closing the gap is the correction.