resume-orchestrator

Resume Orchestrator

You are orchestrating a savage-resume run on a candidate's resume, CV, or portfolio.

Announce at start: "Using resume-orchestrator to run the 8-stage savage-resume pipeline on <target>."

Precondition checks (abort on failure)

1. The argument <target> was provided. If missing → print error: /resume requires a local path or URL argument and stop.
2. Classify <target>:
   • If it starts with http:// or https:// → treat as URL mode.
   • Otherwise → treat as FILE mode. Verify the path exists and is a file (not a directory). If it's a directory, print error: <target> is a directory; pass a single resume file and stop. If it does not exist, print error: <target> does not exist and stop.
3. In FILE mode, verify the extension is one of .pdf, .md, .markdown, .txt, .docx, .html. Unknown extensions → print error: unsupported file type '<ext>'; supported: pdf, md, txt, docx, html and stop. (.docx may require a best-effort text extraction; if unreadable, stage 1 notes it and the pipeline continues on whatever was extractable.)

Setup

1. Compute $TS = current timestamp as YYYY-MM-DD-HHMM (local time).
2. Compute $SUBJECT:
   • FILE mode → basename of the file, stripped of extension (e.g. jane-doe-resume).
   • URL mode → hostname + first path segment, sanitized to ascii + hyphens (e.g. janedoe-com-portfolio).
3. Compute $BASE_DIR:
   • FILE mode → parent directory of the file, absolute.
   • URL mode → current working directory, absolute.
4. Create directories:
   • $BASE_DIR/.resume-review/runs/$TS/scratchpads/
5. The final report path is $BASE_DIR/.resume-review/$TS-$SUBJECT.md.

Stage 1 — Ingest (you perform this directly; no subagent)

Read the target material:

• FILE mode:
  • .md / .markdown / .txt / .html → Read tool.
  • .pdf → Read tool (handles PDFs up to 20 pages per call; if larger, read in page ranges and concatenate).
  • .docx → best-effort: attempt Read; if binary garbage, state the file could not be fully parsed and proceed with whatever was extractable. Record the limitation in the ingest summary.
• URL mode: use WebFetch with a prompt asking for the full visible text content of the page, plus any linked "resume", "CV", "about", or "projects" subpages that are on the same origin and look load-bearing (max 3 extra pages). Record each fetched URL in the ingest summary.

Write a neutral summary to $BASE_DIR/.resume-review/runs/$TS/ingest-summary.md. The summary states, in the subject's own framing:

• Who the candidate says they are (name, headline, one-line positioning)
• Seniority level / role signal (IC, manager, founder, student, career-switcher, etc.)
• Target role(s) or industries, if stated or strongly implied
• Stated skills, tools, and technologies
• Chronological work / project / education history as written (titles, organizations, dates)
• Quantitative claims made anywhere (metrics, scale, team size, revenue, users, percentages)
• Qualitative claims (adjective-bearing self-descriptions: "strategic", "results-driven", "fast learner", etc.)
• Stated differentiators ("unlike most X, I…")
• Overall structural shape of the document (sections, length, layout cues)

Do not interpret, rank, or critique at this stage. Neutral summary only.

If the target is essentially empty (no substantive text), the summary says so explicitly and the pipeline proceeds to a trivial final verdict of "Nothing to evaluate."

Stage 2 — Claim extraction (subagent: claim-extractor)

Dispatch the claim-extractor subagent with the task:

Read $BASE_DIR/.resume-review/runs/$TS/ingest-summary.md and produce a structured claim list at $BASE_DIR/.resume-review/runs/$TS/claims.md conforming to the schema in plugin/skills/claim-extraction/SKILL.md. Also read plugin/skills/resume-rubric/SKILL.md first.

Wait for completion signal claims-extracted: <path>.

Stage 3 — Parallel specialists

Dispatch all five in one batch (parallel tool calls, single message):

• market-demand-hunter → writes $BASE_DIR/.resume-review/runs/$TS/scratchpads/market-demand.md
• claim-auditor → writes $BASE_DIR/.resume-review/runs/$TS/scratchpads/claim-audit.md
• rejection-mode-hunter → writes $BASE_DIR/.resume-review/runs/$TS/scratchpads/rejection-modes.md
• relevance-checker → writes $BASE_DIR/.resume-review/runs/$TS/scratchpads/relevance.md
• positioning-critic → writes $BASE_DIR/.resume-review/runs/$TS/scratchpads/positioning-critique.md

Each gets the same core prompt: "Run your agent on the subject at $BASE_DIR/.resume-review/runs/$TS/. Your inputs are the ingest-summary and claims there. Your output goes to the path named in your agent file. Read plugin/skills/resume-rubric/SKILL.md before producing output."

Wait for all five completion signals.

Stage 4 — Initial synthesis (subagent: synthesizer, stage=4)

Dispatch synthesizer with stage=4:

Read all inputs at $BASE_DIR/.resume-review/runs/$TS/ (ingest-summary, claims, all five scratchpads). Follow plugin/skills/savage-synthesis/SKILL.md Stage 4 rules. Write draft to $BASE_DIR/.resume-review/runs/$TS/draft-v1.md.

Wait for completion.

Stage 5 — Adversary (subagent: adversary)

Dispatch adversary:

Read $BASE_DIR/.resume-review/runs/$TS/draft-v1.md and all scratchpads at $BASE_DIR/.resume-review/runs/$TS/scratchpads/. Produce kill-list at $BASE_DIR/.resume-review/runs/$TS/kill-list.md.

Wait for completion.

Stage 6 — Revision (subagent: synthesizer, stage=6)

Dispatch synthesizer with stage=6:

Read draft-v1.md and kill-list.md at $BASE_DIR/.resume-review/runs/$TS/. Follow Stage 6 rules in savage-synthesis. Write draft-v2.md.

Wait for completion.

Stage 7 — Self-audit (subagent: self-auditor)

Dispatch self-auditor:

Audit $BASE_DIR/.resume-review/runs/$TS/draft-v2.md against kill-list.md. Write audit to $BASE_DIR/.resume-review/runs/$TS/audit-v1.md.

Wait for completion signal audit-result: <PASS|FAIL>.

Branch

• PASS → promote draft-v2.md to final. Go to Stage 8.
• FAIL → dispatch synthesizer with stage=6 again, but this time reading draft-v2.md + audit-v1.md as the kill-list, producing draft-v3.md. Then dispatch self-auditor again producing audit-v2.md.
  • If PASS on audit-v2 → promote draft-v3.md.
  • If FAIL on audit-v2 → promote draft-v3.md anyway, but prepend a ## Remaining Flaws section at the top of the final report copying the Violations list from audit-v2.md.

Stage 8 — Emit

1. Copy the promoted draft to $BASE_DIR/.resume-review/$TS-$SUBJECT.md. If Remaining Flaws were prepended, do that during the copy.

2. Print a terminal summary:

   ============================================================
   Savage resume review complete: <SUBJECT>
   ============================================================
   
   Verdict:
   <paste the Verdict section content>
   
   Top 3 red flags:
   <first 3 bullets from Red flags>
   
   Full report: <absolute path to final report>
   ============================================================
   

Error handling

• Any subagent fails / times out → record the failure in $BASE_DIR/.resume-review/runs/$TS/run.log, and continue if the pipeline can proceed without that input (the synthesizer treats missing scratchpads as "unavailable — this category could not be assessed"). If the synthesizer itself fails, abort and print the run directory path so the user can inspect scratchpads.
• Any web tool fails → the affected specialist records the failure in its own scratchpad; downstream agents handle gracefully (downgrade to Unverifiable).
• Unreadable / corrupt file → Stage 1 produces a summary saying "file could not be parsed"; if nothing extractable, the pipeline short-circuits to a verdict of "Nothing to evaluate."